+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple amdgcn-amd-amdhsa -mcpu=gfx803 -O0 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefix=GCN %s
define void @child_function() #0 {
+; GCN-LABEL: child_function:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: ;;#ASMSTART
+; GCN-NEXT: ;;#ASMEND
+; GCN-NEXT: s_setpc_b64 s[30:31]
call void asm sideeffect "", "~{vcc}" () #0
ret void
}
-; GCN-LABEL: {{^}}spill_sgpr_with_no_lower_vgpr_available:
-; GCN: buffer_store_dword v255, off, s[0:3], s32
-; GCN: v_writelane_b32 v255, s33, 2
-; GCN: v_writelane_b32 v255, s30, 0
-; GCN: v_writelane_b32 v255, s31, 1
-; GCN: s_swappc_b64 s[30:31], s[4:5]
-; GCN: v_readlane_b32 s31, v255, 1
-; GCN: v_readlane_b32 s30, v255, 0
-; GCN: v_readlane_b32 s33, v255, 2
-; GCN: ; NumVgprs: 256
-
define void @spill_sgpr_with_no_lower_vgpr_available() #0 {
+; GCN-LABEL: spill_sgpr_with_no_lower_vgpr_available:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: s_or_saveexec_b64 s[4:5], -1
+; GCN-NEXT: buffer_store_dword v255, off, s[0:3], s32 offset:448 ; 4-byte Folded Spill
+; GCN-NEXT: s_mov_b64 exec, s[4:5]
+; GCN-NEXT: v_writelane_b32 v255, s33, 2
+; GCN-NEXT: s_mov_b32 s33, s32
+; GCN-NEXT: s_add_i32 s32, s32, 0x7400
+; GCN-NEXT: buffer_store_dword v40, off, s[0:3], s33 offset:440 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v41, off, s[0:3], s33 offset:436 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v42, off, s[0:3], s33 offset:432 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v43, off, s[0:3], s33 offset:428 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v44, off, s[0:3], s33 offset:424 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v45, off, s[0:3], s33 offset:420 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v46, off, s[0:3], s33 offset:416 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v47, off, s[0:3], s33 offset:412 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v56, off, s[0:3], s33 offset:408 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v57, off, s[0:3], s33 offset:404 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v58, off, s[0:3], s33 offset:400 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v59, off, s[0:3], s33 offset:396 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v60, off, s[0:3], s33 offset:392 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v61, off, s[0:3], s33 offset:388 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v62, off, s[0:3], s33 offset:384 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v63, off, s[0:3], s33 offset:380 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v72, off, s[0:3], s33 offset:376 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v73, off, s[0:3], s33 offset:372 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v74, off, s[0:3], s33 offset:368 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v75, off, s[0:3], s33 offset:364 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v76, off, s[0:3], s33 offset:360 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v77, off, s[0:3], s33 offset:356 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v78, off, s[0:3], s33 offset:352 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v79, off, s[0:3], s33 offset:348 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v88, off, s[0:3], s33 offset:344 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v89, off, s[0:3], s33 offset:340 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v90, off, s[0:3], s33 offset:336 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v91, off, s[0:3], s33 offset:332 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v92, off, s[0:3], s33 offset:328 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v93, off, s[0:3], s33 offset:324 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v94, off, s[0:3], s33 offset:320 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v95, off, s[0:3], s33 offset:316 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v104, off, s[0:3], s33 offset:312 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v105, off, s[0:3], s33 offset:308 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v106, off, s[0:3], s33 offset:304 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v107, off, s[0:3], s33 offset:300 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v108, off, s[0:3], s33 offset:296 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v109, off, s[0:3], s33 offset:292 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v110, off, s[0:3], s33 offset:288 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v111, off, s[0:3], s33 offset:284 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v120, off, s[0:3], s33 offset:280 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v121, off, s[0:3], s33 offset:276 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v122, off, s[0:3], s33 offset:272 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v123, off, s[0:3], s33 offset:268 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v124, off, s[0:3], s33 offset:264 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v125, off, s[0:3], s33 offset:260 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v126, off, s[0:3], s33 offset:256 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v127, off, s[0:3], s33 offset:252 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v136, off, s[0:3], s33 offset:248 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v137, off, s[0:3], s33 offset:244 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v138, off, s[0:3], s33 offset:240 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v139, off, s[0:3], s33 offset:236 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v140, off, s[0:3], s33 offset:232 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v141, off, s[0:3], s33 offset:228 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v142, off, s[0:3], s33 offset:224 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v143, off, s[0:3], s33 offset:220 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v152, off, s[0:3], s33 offset:216 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v153, off, s[0:3], s33 offset:212 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v154, off, s[0:3], s33 offset:208 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v155, off, s[0:3], s33 offset:204 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v156, off, s[0:3], s33 offset:200 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v157, off, s[0:3], s33 offset:196 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v158, off, s[0:3], s33 offset:192 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v159, off, s[0:3], s33 offset:188 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v168, off, s[0:3], s33 offset:184 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v169, off, s[0:3], s33 offset:180 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v170, off, s[0:3], s33 offset:176 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v171, off, s[0:3], s33 offset:172 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v172, off, s[0:3], s33 offset:168 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v173, off, s[0:3], s33 offset:164 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v174, off, s[0:3], s33 offset:160 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v175, off, s[0:3], s33 offset:156 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v184, off, s[0:3], s33 offset:152 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v185, off, s[0:3], s33 offset:148 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v186, off, s[0:3], s33 offset:144 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v187, off, s[0:3], s33 offset:140 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v188, off, s[0:3], s33 offset:136 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v189, off, s[0:3], s33 offset:132 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v190, off, s[0:3], s33 offset:128 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v191, off, s[0:3], s33 offset:124 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v200, off, s[0:3], s33 offset:120 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v201, off, s[0:3], s33 offset:116 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v202, off, s[0:3], s33 offset:112 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v203, off, s[0:3], s33 offset:108 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v204, off, s[0:3], s33 offset:104 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v205, off, s[0:3], s33 offset:100 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v206, off, s[0:3], s33 offset:96 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v207, off, s[0:3], s33 offset:92 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v216, off, s[0:3], s33 offset:88 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v217, off, s[0:3], s33 offset:84 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v218, off, s[0:3], s33 offset:80 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v219, off, s[0:3], s33 offset:76 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v220, off, s[0:3], s33 offset:72 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v221, off, s[0:3], s33 offset:68 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v222, off, s[0:3], s33 offset:64 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v223, off, s[0:3], s33 offset:60 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v232, off, s[0:3], s33 offset:56 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v233, off, s[0:3], s33 offset:52 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v234, off, s[0:3], s33 offset:48 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v235, off, s[0:3], s33 offset:44 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v236, off, s[0:3], s33 offset:40 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v237, off, s[0:3], s33 offset:36 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v238, off, s[0:3], s33 offset:32 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v239, off, s[0:3], s33 offset:28 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v248, off, s[0:3], s33 offset:24 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v249, off, s[0:3], s33 offset:20 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v250, off, s[0:3], s33 offset:16 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v251, off, s[0:3], s33 offset:12 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v252, off, s[0:3], s33 offset:8 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v253, off, s[0:3], s33 offset:4 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v254, off, s[0:3], s33 ; 4-byte Folded Spill
+; GCN-NEXT: v_writelane_b32 v255, s30, 0
+; GCN-NEXT: v_writelane_b32 v255, s31, 1
+; GCN-NEXT: v_mov_b32_e32 v0, 0
+; GCN-NEXT: buffer_store_dword v0, off, s[0:3], s33 offset:444
+; GCN-NEXT: s_waitcnt vmcnt(0)
+; GCN-NEXT: ;;#ASMSTART
+; GCN-NEXT: ;;#ASMEND
+; GCN-NEXT: s_getpc_b64 s[4:5]
+; GCN-NEXT: s_add_u32 s4, s4, child_function@gotpcrel32@lo+4
+; GCN-NEXT: s_addc_u32 s5, s5, child_function@gotpcrel32@hi+12
+; GCN-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0x0
+; GCN-NEXT: s_mov_b64 s[10:11], s[2:3]
+; GCN-NEXT: s_mov_b64 s[8:9], s[0:1]
+; GCN-NEXT: s_mov_b64 s[0:1], s[8:9]
+; GCN-NEXT: s_mov_b64 s[2:3], s[10:11]
+; GCN-NEXT: s_waitcnt lgkmcnt(0)
+; GCN-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GCN-NEXT: v_readlane_b32 s31, v255, 1
+; GCN-NEXT: v_readlane_b32 s30, v255, 0
+; GCN-NEXT: buffer_load_dword v254, off, s[0:3], s33 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v253, off, s[0:3], s33 offset:4 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v252, off, s[0:3], s33 offset:8 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v251, off, s[0:3], s33 offset:12 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v250, off, s[0:3], s33 offset:16 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v249, off, s[0:3], s33 offset:20 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v248, off, s[0:3], s33 offset:24 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v239, off, s[0:3], s33 offset:28 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v238, off, s[0:3], s33 offset:32 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v237, off, s[0:3], s33 offset:36 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v236, off, s[0:3], s33 offset:40 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v235, off, s[0:3], s33 offset:44 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v234, off, s[0:3], s33 offset:48 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v233, off, s[0:3], s33 offset:52 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v232, off, s[0:3], s33 offset:56 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v223, off, s[0:3], s33 offset:60 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v222, off, s[0:3], s33 offset:64 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v221, off, s[0:3], s33 offset:68 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v220, off, s[0:3], s33 offset:72 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v219, off, s[0:3], s33 offset:76 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v218, off, s[0:3], s33 offset:80 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v217, off, s[0:3], s33 offset:84 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v216, off, s[0:3], s33 offset:88 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v207, off, s[0:3], s33 offset:92 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v206, off, s[0:3], s33 offset:96 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v205, off, s[0:3], s33 offset:100 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v204, off, s[0:3], s33 offset:104 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v203, off, s[0:3], s33 offset:108 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v202, off, s[0:3], s33 offset:112 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v201, off, s[0:3], s33 offset:116 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v200, off, s[0:3], s33 offset:120 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v191, off, s[0:3], s33 offset:124 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v190, off, s[0:3], s33 offset:128 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v189, off, s[0:3], s33 offset:132 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v188, off, s[0:3], s33 offset:136 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v187, off, s[0:3], s33 offset:140 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v186, off, s[0:3], s33 offset:144 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v185, off, s[0:3], s33 offset:148 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v184, off, s[0:3], s33 offset:152 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v175, off, s[0:3], s33 offset:156 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v174, off, s[0:3], s33 offset:160 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v173, off, s[0:3], s33 offset:164 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v172, off, s[0:3], s33 offset:168 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v171, off, s[0:3], s33 offset:172 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v170, off, s[0:3], s33 offset:176 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v169, off, s[0:3], s33 offset:180 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v168, off, s[0:3], s33 offset:184 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v159, off, s[0:3], s33 offset:188 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v158, off, s[0:3], s33 offset:192 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v157, off, s[0:3], s33 offset:196 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v156, off, s[0:3], s33 offset:200 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v155, off, s[0:3], s33 offset:204 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v154, off, s[0:3], s33 offset:208 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v153, off, s[0:3], s33 offset:212 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v152, off, s[0:3], s33 offset:216 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v143, off, s[0:3], s33 offset:220 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v142, off, s[0:3], s33 offset:224 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v141, off, s[0:3], s33 offset:228 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v140, off, s[0:3], s33 offset:232 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v139, off, s[0:3], s33 offset:236 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v138, off, s[0:3], s33 offset:240 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v137, off, s[0:3], s33 offset:244 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v136, off, s[0:3], s33 offset:248 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v127, off, s[0:3], s33 offset:252 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v126, off, s[0:3], s33 offset:256 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v125, off, s[0:3], s33 offset:260 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v124, off, s[0:3], s33 offset:264 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v123, off, s[0:3], s33 offset:268 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v122, off, s[0:3], s33 offset:272 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v121, off, s[0:3], s33 offset:276 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v120, off, s[0:3], s33 offset:280 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v111, off, s[0:3], s33 offset:284 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v110, off, s[0:3], s33 offset:288 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v109, off, s[0:3], s33 offset:292 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v108, off, s[0:3], s33 offset:296 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v107, off, s[0:3], s33 offset:300 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v106, off, s[0:3], s33 offset:304 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v105, off, s[0:3], s33 offset:308 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v104, off, s[0:3], s33 offset:312 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v95, off, s[0:3], s33 offset:316 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v94, off, s[0:3], s33 offset:320 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v93, off, s[0:3], s33 offset:324 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v92, off, s[0:3], s33 offset:328 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v91, off, s[0:3], s33 offset:332 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v90, off, s[0:3], s33 offset:336 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v89, off, s[0:3], s33 offset:340 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v88, off, s[0:3], s33 offset:344 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v79, off, s[0:3], s33 offset:348 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v78, off, s[0:3], s33 offset:352 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v77, off, s[0:3], s33 offset:356 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v76, off, s[0:3], s33 offset:360 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v75, off, s[0:3], s33 offset:364 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v74, off, s[0:3], s33 offset:368 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v73, off, s[0:3], s33 offset:372 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v72, off, s[0:3], s33 offset:376 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v63, off, s[0:3], s33 offset:380 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v62, off, s[0:3], s33 offset:384 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v61, off, s[0:3], s33 offset:388 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v60, off, s[0:3], s33 offset:392 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v59, off, s[0:3], s33 offset:396 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v58, off, s[0:3], s33 offset:400 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v57, off, s[0:3], s33 offset:404 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v56, off, s[0:3], s33 offset:408 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v47, off, s[0:3], s33 offset:412 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v46, off, s[0:3], s33 offset:416 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v45, off, s[0:3], s33 offset:420 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v44, off, s[0:3], s33 offset:424 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v43, off, s[0:3], s33 offset:428 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v42, off, s[0:3], s33 offset:432 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v41, off, s[0:3], s33 offset:436 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v40, off, s[0:3], s33 offset:440 ; 4-byte Folded Reload
+; GCN-NEXT: s_add_i32 s32, s32, 0xffff8c00
+; GCN-NEXT: v_readlane_b32 s33, v255, 2
+; GCN-NEXT: s_or_saveexec_b64 s[4:5], -1
+; GCN-NEXT: buffer_load_dword v255, off, s[0:3], s32 offset:448 ; 4-byte Folded Reload
+; GCN-NEXT: s_mov_b64 exec, s[4:5]
+; GCN-NEXT: s_waitcnt vmcnt(0)
+; GCN-NEXT: s_setpc_b64 s[30:31]
%alloca = alloca i32, align 4, addrspace(5)
store volatile i32 0, i32 addrspace(5)* %alloca
ret void
}
-; GCN-LABEL: {{^}}spill_to_lowest_available_vgpr:
-; GCN: buffer_store_dword v254, off, s[0:3], s32
-; GCN: v_writelane_b32 v254, s33, 2
-; GCN: v_writelane_b32 v254, s30, 0
-; GCN: v_writelane_b32 v254, s31, 1
-; GCN: s_swappc_b64 s[30:31], s[4:5]
-; GCN: v_readlane_b32 s31, v254, 1
-; GCN: v_readlane_b32 s30, v254, 0
-; GCN: v_readlane_b32 s33, v254, 2
-
define void @spill_to_lowest_available_vgpr() #0 {
+; GCN-LABEL: spill_to_lowest_available_vgpr:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: s_or_saveexec_b64 s[4:5], -1
+; GCN-NEXT: buffer_store_dword v254, off, s[0:3], s32 offset:444 ; 4-byte Folded Spill
+; GCN-NEXT: s_mov_b64 exec, s[4:5]
+; GCN-NEXT: v_writelane_b32 v254, s33, 2
+; GCN-NEXT: s_mov_b32 s33, s32
+; GCN-NEXT: s_add_i32 s32, s32, 0x7400
+; GCN-NEXT: buffer_store_dword v40, off, s[0:3], s33 offset:436 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v41, off, s[0:3], s33 offset:432 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v42, off, s[0:3], s33 offset:428 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v43, off, s[0:3], s33 offset:424 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v44, off, s[0:3], s33 offset:420 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v45, off, s[0:3], s33 offset:416 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v46, off, s[0:3], s33 offset:412 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v47, off, s[0:3], s33 offset:408 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v56, off, s[0:3], s33 offset:404 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v57, off, s[0:3], s33 offset:400 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v58, off, s[0:3], s33 offset:396 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v59, off, s[0:3], s33 offset:392 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v60, off, s[0:3], s33 offset:388 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v61, off, s[0:3], s33 offset:384 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v62, off, s[0:3], s33 offset:380 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v63, off, s[0:3], s33 offset:376 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v72, off, s[0:3], s33 offset:372 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v73, off, s[0:3], s33 offset:368 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v74, off, s[0:3], s33 offset:364 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v75, off, s[0:3], s33 offset:360 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v76, off, s[0:3], s33 offset:356 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v77, off, s[0:3], s33 offset:352 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v78, off, s[0:3], s33 offset:348 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v79, off, s[0:3], s33 offset:344 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v88, off, s[0:3], s33 offset:340 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v89, off, s[0:3], s33 offset:336 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v90, off, s[0:3], s33 offset:332 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v91, off, s[0:3], s33 offset:328 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v92, off, s[0:3], s33 offset:324 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v93, off, s[0:3], s33 offset:320 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v94, off, s[0:3], s33 offset:316 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v95, off, s[0:3], s33 offset:312 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v104, off, s[0:3], s33 offset:308 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v105, off, s[0:3], s33 offset:304 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v106, off, s[0:3], s33 offset:300 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v107, off, s[0:3], s33 offset:296 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v108, off, s[0:3], s33 offset:292 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v109, off, s[0:3], s33 offset:288 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v110, off, s[0:3], s33 offset:284 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v111, off, s[0:3], s33 offset:280 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v120, off, s[0:3], s33 offset:276 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v121, off, s[0:3], s33 offset:272 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v122, off, s[0:3], s33 offset:268 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v123, off, s[0:3], s33 offset:264 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v124, off, s[0:3], s33 offset:260 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v125, off, s[0:3], s33 offset:256 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v126, off, s[0:3], s33 offset:252 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v127, off, s[0:3], s33 offset:248 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v136, off, s[0:3], s33 offset:244 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v137, off, s[0:3], s33 offset:240 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v138, off, s[0:3], s33 offset:236 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v139, off, s[0:3], s33 offset:232 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v140, off, s[0:3], s33 offset:228 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v141, off, s[0:3], s33 offset:224 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v142, off, s[0:3], s33 offset:220 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v143, off, s[0:3], s33 offset:216 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v152, off, s[0:3], s33 offset:212 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v153, off, s[0:3], s33 offset:208 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v154, off, s[0:3], s33 offset:204 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v155, off, s[0:3], s33 offset:200 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v156, off, s[0:3], s33 offset:196 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v157, off, s[0:3], s33 offset:192 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v158, off, s[0:3], s33 offset:188 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v159, off, s[0:3], s33 offset:184 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v168, off, s[0:3], s33 offset:180 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v169, off, s[0:3], s33 offset:176 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v170, off, s[0:3], s33 offset:172 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v171, off, s[0:3], s33 offset:168 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v172, off, s[0:3], s33 offset:164 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v173, off, s[0:3], s33 offset:160 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v174, off, s[0:3], s33 offset:156 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v175, off, s[0:3], s33 offset:152 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v184, off, s[0:3], s33 offset:148 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v185, off, s[0:3], s33 offset:144 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v186, off, s[0:3], s33 offset:140 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v187, off, s[0:3], s33 offset:136 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v188, off, s[0:3], s33 offset:132 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v189, off, s[0:3], s33 offset:128 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v190, off, s[0:3], s33 offset:124 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v191, off, s[0:3], s33 offset:120 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v200, off, s[0:3], s33 offset:116 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v201, off, s[0:3], s33 offset:112 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v202, off, s[0:3], s33 offset:108 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v203, off, s[0:3], s33 offset:104 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v204, off, s[0:3], s33 offset:100 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v205, off, s[0:3], s33 offset:96 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v206, off, s[0:3], s33 offset:92 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v207, off, s[0:3], s33 offset:88 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v216, off, s[0:3], s33 offset:84 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v217, off, s[0:3], s33 offset:80 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v218, off, s[0:3], s33 offset:76 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v219, off, s[0:3], s33 offset:72 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v220, off, s[0:3], s33 offset:68 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v221, off, s[0:3], s33 offset:64 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v222, off, s[0:3], s33 offset:60 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v223, off, s[0:3], s33 offset:56 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v232, off, s[0:3], s33 offset:52 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v233, off, s[0:3], s33 offset:48 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v234, off, s[0:3], s33 offset:44 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v235, off, s[0:3], s33 offset:40 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v236, off, s[0:3], s33 offset:36 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v237, off, s[0:3], s33 offset:32 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v238, off, s[0:3], s33 offset:28 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v239, off, s[0:3], s33 offset:24 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v248, off, s[0:3], s33 offset:20 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v249, off, s[0:3], s33 offset:16 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v250, off, s[0:3], s33 offset:12 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v251, off, s[0:3], s33 offset:8 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v252, off, s[0:3], s33 offset:4 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v253, off, s[0:3], s33 ; 4-byte Folded Spill
+; GCN-NEXT: v_writelane_b32 v254, s30, 0
+; GCN-NEXT: v_writelane_b32 v254, s31, 1
+; GCN-NEXT: v_mov_b32_e32 v0, 0
+; GCN-NEXT: buffer_store_dword v0, off, s[0:3], s33 offset:440
+; GCN-NEXT: s_waitcnt vmcnt(0)
+; GCN-NEXT: ;;#ASMSTART
+; GCN-NEXT: ;;#ASMEND
+; GCN-NEXT: s_getpc_b64 s[4:5]
+; GCN-NEXT: s_add_u32 s4, s4, child_function@gotpcrel32@lo+4
+; GCN-NEXT: s_addc_u32 s5, s5, child_function@gotpcrel32@hi+12
+; GCN-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0x0
+; GCN-NEXT: s_mov_b64 s[10:11], s[2:3]
+; GCN-NEXT: s_mov_b64 s[8:9], s[0:1]
+; GCN-NEXT: s_mov_b64 s[0:1], s[8:9]
+; GCN-NEXT: s_mov_b64 s[2:3], s[10:11]
+; GCN-NEXT: s_waitcnt lgkmcnt(0)
+; GCN-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GCN-NEXT: v_readlane_b32 s31, v254, 1
+; GCN-NEXT: v_readlane_b32 s30, v254, 0
+; GCN-NEXT: buffer_load_dword v253, off, s[0:3], s33 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v252, off, s[0:3], s33 offset:4 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v251, off, s[0:3], s33 offset:8 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v250, off, s[0:3], s33 offset:12 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v249, off, s[0:3], s33 offset:16 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v248, off, s[0:3], s33 offset:20 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v239, off, s[0:3], s33 offset:24 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v238, off, s[0:3], s33 offset:28 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v237, off, s[0:3], s33 offset:32 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v236, off, s[0:3], s33 offset:36 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v235, off, s[0:3], s33 offset:40 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v234, off, s[0:3], s33 offset:44 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v233, off, s[0:3], s33 offset:48 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v232, off, s[0:3], s33 offset:52 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v223, off, s[0:3], s33 offset:56 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v222, off, s[0:3], s33 offset:60 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v221, off, s[0:3], s33 offset:64 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v220, off, s[0:3], s33 offset:68 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v219, off, s[0:3], s33 offset:72 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v218, off, s[0:3], s33 offset:76 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v217, off, s[0:3], s33 offset:80 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v216, off, s[0:3], s33 offset:84 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v207, off, s[0:3], s33 offset:88 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v206, off, s[0:3], s33 offset:92 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v205, off, s[0:3], s33 offset:96 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v204, off, s[0:3], s33 offset:100 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v203, off, s[0:3], s33 offset:104 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v202, off, s[0:3], s33 offset:108 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v201, off, s[0:3], s33 offset:112 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v200, off, s[0:3], s33 offset:116 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v191, off, s[0:3], s33 offset:120 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v190, off, s[0:3], s33 offset:124 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v189, off, s[0:3], s33 offset:128 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v188, off, s[0:3], s33 offset:132 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v187, off, s[0:3], s33 offset:136 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v186, off, s[0:3], s33 offset:140 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v185, off, s[0:3], s33 offset:144 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v184, off, s[0:3], s33 offset:148 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v175, off, s[0:3], s33 offset:152 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v174, off, s[0:3], s33 offset:156 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v173, off, s[0:3], s33 offset:160 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v172, off, s[0:3], s33 offset:164 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v171, off, s[0:3], s33 offset:168 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v170, off, s[0:3], s33 offset:172 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v169, off, s[0:3], s33 offset:176 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v168, off, s[0:3], s33 offset:180 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v159, off, s[0:3], s33 offset:184 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v158, off, s[0:3], s33 offset:188 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v157, off, s[0:3], s33 offset:192 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v156, off, s[0:3], s33 offset:196 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v155, off, s[0:3], s33 offset:200 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v154, off, s[0:3], s33 offset:204 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v153, off, s[0:3], s33 offset:208 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v152, off, s[0:3], s33 offset:212 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v143, off, s[0:3], s33 offset:216 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v142, off, s[0:3], s33 offset:220 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v141, off, s[0:3], s33 offset:224 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v140, off, s[0:3], s33 offset:228 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v139, off, s[0:3], s33 offset:232 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v138, off, s[0:3], s33 offset:236 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v137, off, s[0:3], s33 offset:240 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v136, off, s[0:3], s33 offset:244 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v127, off, s[0:3], s33 offset:248 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v126, off, s[0:3], s33 offset:252 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v125, off, s[0:3], s33 offset:256 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v124, off, s[0:3], s33 offset:260 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v123, off, s[0:3], s33 offset:264 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v122, off, s[0:3], s33 offset:268 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v121, off, s[0:3], s33 offset:272 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v120, off, s[0:3], s33 offset:276 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v111, off, s[0:3], s33 offset:280 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v110, off, s[0:3], s33 offset:284 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v109, off, s[0:3], s33 offset:288 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v108, off, s[0:3], s33 offset:292 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v107, off, s[0:3], s33 offset:296 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v106, off, s[0:3], s33 offset:300 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v105, off, s[0:3], s33 offset:304 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v104, off, s[0:3], s33 offset:308 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v95, off, s[0:3], s33 offset:312 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v94, off, s[0:3], s33 offset:316 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v93, off, s[0:3], s33 offset:320 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v92, off, s[0:3], s33 offset:324 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v91, off, s[0:3], s33 offset:328 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v90, off, s[0:3], s33 offset:332 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v89, off, s[0:3], s33 offset:336 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v88, off, s[0:3], s33 offset:340 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v79, off, s[0:3], s33 offset:344 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v78, off, s[0:3], s33 offset:348 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v77, off, s[0:3], s33 offset:352 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v76, off, s[0:3], s33 offset:356 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v75, off, s[0:3], s33 offset:360 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v74, off, s[0:3], s33 offset:364 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v73, off, s[0:3], s33 offset:368 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v72, off, s[0:3], s33 offset:372 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v63, off, s[0:3], s33 offset:376 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v62, off, s[0:3], s33 offset:380 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v61, off, s[0:3], s33 offset:384 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v60, off, s[0:3], s33 offset:388 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v59, off, s[0:3], s33 offset:392 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v58, off, s[0:3], s33 offset:396 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v57, off, s[0:3], s33 offset:400 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v56, off, s[0:3], s33 offset:404 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v47, off, s[0:3], s33 offset:408 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v46, off, s[0:3], s33 offset:412 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v45, off, s[0:3], s33 offset:416 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v44, off, s[0:3], s33 offset:420 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v43, off, s[0:3], s33 offset:424 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v42, off, s[0:3], s33 offset:428 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v41, off, s[0:3], s33 offset:432 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v40, off, s[0:3], s33 offset:436 ; 4-byte Folded Reload
+; GCN-NEXT: s_add_i32 s32, s32, 0xffff8c00
+; GCN-NEXT: v_readlane_b32 s33, v254, 2
+; GCN-NEXT: s_or_saveexec_b64 s[4:5], -1
+; GCN-NEXT: buffer_load_dword v254, off, s[0:3], s32 offset:444 ; 4-byte Folded Reload
+; GCN-NEXT: s_mov_b64 exec, s[4:5]
+; GCN-NEXT: s_waitcnt vmcnt(0)
+; GCN-NEXT: s_setpc_b64 s[30:31]
%alloca = alloca i32, align 4, addrspace(5)
store volatile i32 0, i32 addrspace(5)* %alloca
ret void
}
-; GCN-LABEL: {{^}}spill_sgpr_with_sgpr_uses:
-; GCN-NOT: buffer_store_dword v255, off, s[0:3], s32
-; GCN: ; def s4
-; GCN: v_writelane_b32 v254, s4, 0
-; GCN: v_readlane_b32 s4, v254, 0
-; GCN: ; use s4
-
define void @spill_sgpr_with_sgpr_uses() #0 {
+; GCN-LABEL: spill_sgpr_with_sgpr_uses:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: s_or_saveexec_b64 s[4:5], -1
+; GCN-NEXT: buffer_store_dword v254, off, s[0:3], s32 offset:444 ; 4-byte Folded Spill
+; GCN-NEXT: s_mov_b64 exec, s[4:5]
+; GCN-NEXT: buffer_store_dword v40, off, s[0:3], s32 offset:436 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v41, off, s[0:3], s32 offset:432 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v42, off, s[0:3], s32 offset:428 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v43, off, s[0:3], s32 offset:424 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v44, off, s[0:3], s32 offset:420 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v45, off, s[0:3], s32 offset:416 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v46, off, s[0:3], s32 offset:412 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v47, off, s[0:3], s32 offset:408 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v56, off, s[0:3], s32 offset:404 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v57, off, s[0:3], s32 offset:400 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v58, off, s[0:3], s32 offset:396 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v59, off, s[0:3], s32 offset:392 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v60, off, s[0:3], s32 offset:388 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v61, off, s[0:3], s32 offset:384 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v62, off, s[0:3], s32 offset:380 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v63, off, s[0:3], s32 offset:376 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v72, off, s[0:3], s32 offset:372 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v73, off, s[0:3], s32 offset:368 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v74, off, s[0:3], s32 offset:364 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v75, off, s[0:3], s32 offset:360 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v76, off, s[0:3], s32 offset:356 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v77, off, s[0:3], s32 offset:352 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v78, off, s[0:3], s32 offset:348 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v79, off, s[0:3], s32 offset:344 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v88, off, s[0:3], s32 offset:340 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v89, off, s[0:3], s32 offset:336 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v90, off, s[0:3], s32 offset:332 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v91, off, s[0:3], s32 offset:328 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v92, off, s[0:3], s32 offset:324 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v93, off, s[0:3], s32 offset:320 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v94, off, s[0:3], s32 offset:316 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v95, off, s[0:3], s32 offset:312 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v104, off, s[0:3], s32 offset:308 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v105, off, s[0:3], s32 offset:304 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v106, off, s[0:3], s32 offset:300 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v107, off, s[0:3], s32 offset:296 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v108, off, s[0:3], s32 offset:292 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v109, off, s[0:3], s32 offset:288 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v110, off, s[0:3], s32 offset:284 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v111, off, s[0:3], s32 offset:280 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v120, off, s[0:3], s32 offset:276 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v121, off, s[0:3], s32 offset:272 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v122, off, s[0:3], s32 offset:268 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v123, off, s[0:3], s32 offset:264 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v124, off, s[0:3], s32 offset:260 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v125, off, s[0:3], s32 offset:256 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v126, off, s[0:3], s32 offset:252 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v127, off, s[0:3], s32 offset:248 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v136, off, s[0:3], s32 offset:244 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v137, off, s[0:3], s32 offset:240 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v138, off, s[0:3], s32 offset:236 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v139, off, s[0:3], s32 offset:232 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v140, off, s[0:3], s32 offset:228 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v141, off, s[0:3], s32 offset:224 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v142, off, s[0:3], s32 offset:220 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v143, off, s[0:3], s32 offset:216 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v152, off, s[0:3], s32 offset:212 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v153, off, s[0:3], s32 offset:208 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v154, off, s[0:3], s32 offset:204 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v155, off, s[0:3], s32 offset:200 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v156, off, s[0:3], s32 offset:196 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v157, off, s[0:3], s32 offset:192 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v158, off, s[0:3], s32 offset:188 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v159, off, s[0:3], s32 offset:184 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v168, off, s[0:3], s32 offset:180 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v169, off, s[0:3], s32 offset:176 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v170, off, s[0:3], s32 offset:172 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v171, off, s[0:3], s32 offset:168 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v172, off, s[0:3], s32 offset:164 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v173, off, s[0:3], s32 offset:160 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v174, off, s[0:3], s32 offset:156 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v175, off, s[0:3], s32 offset:152 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v184, off, s[0:3], s32 offset:148 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v185, off, s[0:3], s32 offset:144 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v186, off, s[0:3], s32 offset:140 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v187, off, s[0:3], s32 offset:136 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v188, off, s[0:3], s32 offset:132 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v189, off, s[0:3], s32 offset:128 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v190, off, s[0:3], s32 offset:124 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v191, off, s[0:3], s32 offset:120 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v200, off, s[0:3], s32 offset:116 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v201, off, s[0:3], s32 offset:112 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v202, off, s[0:3], s32 offset:108 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v203, off, s[0:3], s32 offset:104 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v204, off, s[0:3], s32 offset:100 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v205, off, s[0:3], s32 offset:96 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v206, off, s[0:3], s32 offset:92 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v207, off, s[0:3], s32 offset:88 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v216, off, s[0:3], s32 offset:84 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v217, off, s[0:3], s32 offset:80 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v218, off, s[0:3], s32 offset:76 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v219, off, s[0:3], s32 offset:72 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v220, off, s[0:3], s32 offset:68 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v221, off, s[0:3], s32 offset:64 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v222, off, s[0:3], s32 offset:60 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v223, off, s[0:3], s32 offset:56 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v232, off, s[0:3], s32 offset:52 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v233, off, s[0:3], s32 offset:48 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v234, off, s[0:3], s32 offset:44 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v235, off, s[0:3], s32 offset:40 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v236, off, s[0:3], s32 offset:36 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v237, off, s[0:3], s32 offset:32 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v238, off, s[0:3], s32 offset:28 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v239, off, s[0:3], s32 offset:24 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v248, off, s[0:3], s32 offset:20 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v249, off, s[0:3], s32 offset:16 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v250, off, s[0:3], s32 offset:12 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v251, off, s[0:3], s32 offset:8 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v252, off, s[0:3], s32 offset:4 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v253, off, s[0:3], s32 ; 4-byte Folded Spill
+; GCN-NEXT: v_mov_b32_e32 v0, 0
+; GCN-NEXT: buffer_store_dword v0, off, s[0:3], s32 offset:440
+; GCN-NEXT: s_waitcnt vmcnt(0)
+; GCN-NEXT: ;;#ASMSTART
+; GCN-NEXT: ;;#ASMEND
+; GCN-NEXT: ;;#ASMSTART
+; GCN-NEXT: ; def s4
+; GCN-NEXT: ;;#ASMEND
+; GCN-NEXT: v_writelane_b32 v254, s4, 0
+; GCN-NEXT: s_cbranch_scc1 .LBB3_2
+; GCN-NEXT: ; %bb.1: ; %bb0
+; GCN-NEXT: v_readlane_b32 s4, v254, 0
+; GCN-NEXT: ;;#ASMSTART
+; GCN-NEXT: ; use s4
+; GCN-NEXT: ;;#ASMEND
+; GCN-NEXT: .LBB3_2: ; %ret
+; GCN-NEXT: buffer_load_dword v253, off, s[0:3], s32 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v252, off, s[0:3], s32 offset:4 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v251, off, s[0:3], s32 offset:8 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v250, off, s[0:3], s32 offset:12 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v249, off, s[0:3], s32 offset:16 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v248, off, s[0:3], s32 offset:20 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v239, off, s[0:3], s32 offset:24 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v238, off, s[0:3], s32 offset:28 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v237, off, s[0:3], s32 offset:32 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v236, off, s[0:3], s32 offset:36 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v235, off, s[0:3], s32 offset:40 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v234, off, s[0:3], s32 offset:44 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v233, off, s[0:3], s32 offset:48 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v232, off, s[0:3], s32 offset:52 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v223, off, s[0:3], s32 offset:56 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v222, off, s[0:3], s32 offset:60 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v221, off, s[0:3], s32 offset:64 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v220, off, s[0:3], s32 offset:68 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v219, off, s[0:3], s32 offset:72 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v218, off, s[0:3], s32 offset:76 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v217, off, s[0:3], s32 offset:80 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v216, off, s[0:3], s32 offset:84 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v207, off, s[0:3], s32 offset:88 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v206, off, s[0:3], s32 offset:92 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v205, off, s[0:3], s32 offset:96 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v204, off, s[0:3], s32 offset:100 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v203, off, s[0:3], s32 offset:104 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v202, off, s[0:3], s32 offset:108 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v201, off, s[0:3], s32 offset:112 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v200, off, s[0:3], s32 offset:116 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v191, off, s[0:3], s32 offset:120 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v190, off, s[0:3], s32 offset:124 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v189, off, s[0:3], s32 offset:128 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v188, off, s[0:3], s32 offset:132 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v187, off, s[0:3], s32 offset:136 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v186, off, s[0:3], s32 offset:140 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v185, off, s[0:3], s32 offset:144 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v184, off, s[0:3], s32 offset:148 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v175, off, s[0:3], s32 offset:152 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v174, off, s[0:3], s32 offset:156 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v173, off, s[0:3], s32 offset:160 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v172, off, s[0:3], s32 offset:164 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v171, off, s[0:3], s32 offset:168 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v170, off, s[0:3], s32 offset:172 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v169, off, s[0:3], s32 offset:176 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v168, off, s[0:3], s32 offset:180 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v159, off, s[0:3], s32 offset:184 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v158, off, s[0:3], s32 offset:188 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v157, off, s[0:3], s32 offset:192 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v156, off, s[0:3], s32 offset:196 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v155, off, s[0:3], s32 offset:200 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v154, off, s[0:3], s32 offset:204 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v153, off, s[0:3], s32 offset:208 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v152, off, s[0:3], s32 offset:212 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v143, off, s[0:3], s32 offset:216 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v142, off, s[0:3], s32 offset:220 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v141, off, s[0:3], s32 offset:224 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v140, off, s[0:3], s32 offset:228 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v139, off, s[0:3], s32 offset:232 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v138, off, s[0:3], s32 offset:236 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v137, off, s[0:3], s32 offset:240 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v136, off, s[0:3], s32 offset:244 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v127, off, s[0:3], s32 offset:248 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v126, off, s[0:3], s32 offset:252 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v125, off, s[0:3], s32 offset:256 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v124, off, s[0:3], s32 offset:260 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v123, off, s[0:3], s32 offset:264 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v122, off, s[0:3], s32 offset:268 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v121, off, s[0:3], s32 offset:272 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v120, off, s[0:3], s32 offset:276 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v111, off, s[0:3], s32 offset:280 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v110, off, s[0:3], s32 offset:284 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v109, off, s[0:3], s32 offset:288 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v108, off, s[0:3], s32 offset:292 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v107, off, s[0:3], s32 offset:296 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v106, off, s[0:3], s32 offset:300 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v105, off, s[0:3], s32 offset:304 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v104, off, s[0:3], s32 offset:308 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v95, off, s[0:3], s32 offset:312 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v94, off, s[0:3], s32 offset:316 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v93, off, s[0:3], s32 offset:320 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v92, off, s[0:3], s32 offset:324 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v91, off, s[0:3], s32 offset:328 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v90, off, s[0:3], s32 offset:332 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v89, off, s[0:3], s32 offset:336 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v88, off, s[0:3], s32 offset:340 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v79, off, s[0:3], s32 offset:344 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v78, off, s[0:3], s32 offset:348 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v77, off, s[0:3], s32 offset:352 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v76, off, s[0:3], s32 offset:356 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v75, off, s[0:3], s32 offset:360 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v74, off, s[0:3], s32 offset:364 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v73, off, s[0:3], s32 offset:368 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v72, off, s[0:3], s32 offset:372 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v63, off, s[0:3], s32 offset:376 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v62, off, s[0:3], s32 offset:380 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v61, off, s[0:3], s32 offset:384 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v60, off, s[0:3], s32 offset:388 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v59, off, s[0:3], s32 offset:392 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v58, off, s[0:3], s32 offset:396 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v57, off, s[0:3], s32 offset:400 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v56, off, s[0:3], s32 offset:404 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v47, off, s[0:3], s32 offset:408 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v46, off, s[0:3], s32 offset:412 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v45, off, s[0:3], s32 offset:416 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v44, off, s[0:3], s32 offset:420 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v43, off, s[0:3], s32 offset:424 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v42, off, s[0:3], s32 offset:428 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v41, off, s[0:3], s32 offset:432 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v40, off, s[0:3], s32 offset:436 ; 4-byte Folded Reload
+; GCN-NEXT: s_or_saveexec_b64 s[4:5], -1
+; GCN-NEXT: buffer_load_dword v254, off, s[0:3], s32 offset:444 ; 4-byte Folded Reload
+; GCN-NEXT: s_mov_b64 exec, s[4:5]
+; GCN-NEXT: s_waitcnt vmcnt(0)
+; GCN-NEXT: s_setpc_b64 s[30:31]
%alloca = alloca i32, align 4, addrspace(5)
store volatile i32 0, i32 addrspace(5)* %alloca
ret void
}
-; GCN-LABEL: {{^}}spill_sgpr_with_tail_call
-; GCN-NOT: buffer_store_dword v255, off, s[0:3], s32
-; GCN-NOT: v_writelane
-; GCN: s_setpc_b64 s[4:5]
-
define void @spill_sgpr_with_tail_call() #0 {
+; GCN-LABEL: spill_sgpr_with_tail_call:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: buffer_store_dword v40, off, s[0:3], s32 offset:440 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v41, off, s[0:3], s32 offset:436 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v42, off, s[0:3], s32 offset:432 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v43, off, s[0:3], s32 offset:428 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v44, off, s[0:3], s32 offset:424 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v45, off, s[0:3], s32 offset:420 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v46, off, s[0:3], s32 offset:416 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v47, off, s[0:3], s32 offset:412 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v56, off, s[0:3], s32 offset:408 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v57, off, s[0:3], s32 offset:404 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v58, off, s[0:3], s32 offset:400 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v59, off, s[0:3], s32 offset:396 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v60, off, s[0:3], s32 offset:392 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v61, off, s[0:3], s32 offset:388 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v62, off, s[0:3], s32 offset:384 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v63, off, s[0:3], s32 offset:380 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v72, off, s[0:3], s32 offset:376 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v73, off, s[0:3], s32 offset:372 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v74, off, s[0:3], s32 offset:368 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v75, off, s[0:3], s32 offset:364 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v76, off, s[0:3], s32 offset:360 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v77, off, s[0:3], s32 offset:356 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v78, off, s[0:3], s32 offset:352 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v79, off, s[0:3], s32 offset:348 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v88, off, s[0:3], s32 offset:344 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v89, off, s[0:3], s32 offset:340 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v90, off, s[0:3], s32 offset:336 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v91, off, s[0:3], s32 offset:332 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v92, off, s[0:3], s32 offset:328 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v93, off, s[0:3], s32 offset:324 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v94, off, s[0:3], s32 offset:320 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v95, off, s[0:3], s32 offset:316 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v104, off, s[0:3], s32 offset:312 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v105, off, s[0:3], s32 offset:308 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v106, off, s[0:3], s32 offset:304 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v107, off, s[0:3], s32 offset:300 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v108, off, s[0:3], s32 offset:296 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v109, off, s[0:3], s32 offset:292 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v110, off, s[0:3], s32 offset:288 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v111, off, s[0:3], s32 offset:284 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v120, off, s[0:3], s32 offset:280 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v121, off, s[0:3], s32 offset:276 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v122, off, s[0:3], s32 offset:272 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v123, off, s[0:3], s32 offset:268 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v124, off, s[0:3], s32 offset:264 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v125, off, s[0:3], s32 offset:260 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v126, off, s[0:3], s32 offset:256 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v127, off, s[0:3], s32 offset:252 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v136, off, s[0:3], s32 offset:248 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v137, off, s[0:3], s32 offset:244 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v138, off, s[0:3], s32 offset:240 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v139, off, s[0:3], s32 offset:236 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v140, off, s[0:3], s32 offset:232 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v141, off, s[0:3], s32 offset:228 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v142, off, s[0:3], s32 offset:224 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v143, off, s[0:3], s32 offset:220 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v152, off, s[0:3], s32 offset:216 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v153, off, s[0:3], s32 offset:212 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v154, off, s[0:3], s32 offset:208 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v155, off, s[0:3], s32 offset:204 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v156, off, s[0:3], s32 offset:200 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v157, off, s[0:3], s32 offset:196 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v158, off, s[0:3], s32 offset:192 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v159, off, s[0:3], s32 offset:188 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v168, off, s[0:3], s32 offset:184 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v169, off, s[0:3], s32 offset:180 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v170, off, s[0:3], s32 offset:176 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v171, off, s[0:3], s32 offset:172 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v172, off, s[0:3], s32 offset:168 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v173, off, s[0:3], s32 offset:164 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v174, off, s[0:3], s32 offset:160 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v175, off, s[0:3], s32 offset:156 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v184, off, s[0:3], s32 offset:152 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v185, off, s[0:3], s32 offset:148 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v186, off, s[0:3], s32 offset:144 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v187, off, s[0:3], s32 offset:140 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v188, off, s[0:3], s32 offset:136 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v189, off, s[0:3], s32 offset:132 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v190, off, s[0:3], s32 offset:128 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v191, off, s[0:3], s32 offset:124 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v200, off, s[0:3], s32 offset:120 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v201, off, s[0:3], s32 offset:116 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v202, off, s[0:3], s32 offset:112 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v203, off, s[0:3], s32 offset:108 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v204, off, s[0:3], s32 offset:104 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v205, off, s[0:3], s32 offset:100 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v206, off, s[0:3], s32 offset:96 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v207, off, s[0:3], s32 offset:92 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v216, off, s[0:3], s32 offset:88 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v217, off, s[0:3], s32 offset:84 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v218, off, s[0:3], s32 offset:80 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v219, off, s[0:3], s32 offset:76 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v220, off, s[0:3], s32 offset:72 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v221, off, s[0:3], s32 offset:68 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v222, off, s[0:3], s32 offset:64 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v223, off, s[0:3], s32 offset:60 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v232, off, s[0:3], s32 offset:56 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v233, off, s[0:3], s32 offset:52 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v234, off, s[0:3], s32 offset:48 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v235, off, s[0:3], s32 offset:44 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v236, off, s[0:3], s32 offset:40 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v237, off, s[0:3], s32 offset:36 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v238, off, s[0:3], s32 offset:32 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v239, off, s[0:3], s32 offset:28 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v248, off, s[0:3], s32 offset:24 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v249, off, s[0:3], s32 offset:20 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v250, off, s[0:3], s32 offset:16 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v251, off, s[0:3], s32 offset:12 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v252, off, s[0:3], s32 offset:8 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v253, off, s[0:3], s32 offset:4 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v254, off, s[0:3], s32 ; 4-byte Folded Spill
+; GCN-NEXT: v_mov_b32_e32 v0, 0
+; GCN-NEXT: buffer_store_dword v0, off, s[0:3], s32 offset:444
+; GCN-NEXT: s_waitcnt vmcnt(0)
+; GCN-NEXT: ;;#ASMSTART
+; GCN-NEXT: ;;#ASMEND
+; GCN-NEXT: s_getpc_b64 s[4:5]
+; GCN-NEXT: s_add_u32 s4, s4, child_function@gotpcrel32@lo+4
+; GCN-NEXT: s_addc_u32 s5, s5, child_function@gotpcrel32@hi+12
+; GCN-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0x0
+; GCN-NEXT: buffer_load_dword v254, off, s[0:3], s32 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v253, off, s[0:3], s32 offset:4 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v252, off, s[0:3], s32 offset:8 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v251, off, s[0:3], s32 offset:12 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v250, off, s[0:3], s32 offset:16 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v249, off, s[0:3], s32 offset:20 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v248, off, s[0:3], s32 offset:24 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v239, off, s[0:3], s32 offset:28 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v238, off, s[0:3], s32 offset:32 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v237, off, s[0:3], s32 offset:36 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v236, off, s[0:3], s32 offset:40 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v235, off, s[0:3], s32 offset:44 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v234, off, s[0:3], s32 offset:48 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v233, off, s[0:3], s32 offset:52 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v232, off, s[0:3], s32 offset:56 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v223, off, s[0:3], s32 offset:60 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v222, off, s[0:3], s32 offset:64 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v221, off, s[0:3], s32 offset:68 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v220, off, s[0:3], s32 offset:72 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v219, off, s[0:3], s32 offset:76 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v218, off, s[0:3], s32 offset:80 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v217, off, s[0:3], s32 offset:84 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v216, off, s[0:3], s32 offset:88 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v207, off, s[0:3], s32 offset:92 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v206, off, s[0:3], s32 offset:96 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v205, off, s[0:3], s32 offset:100 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v204, off, s[0:3], s32 offset:104 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v203, off, s[0:3], s32 offset:108 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v202, off, s[0:3], s32 offset:112 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v201, off, s[0:3], s32 offset:116 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v200, off, s[0:3], s32 offset:120 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v191, off, s[0:3], s32 offset:124 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v190, off, s[0:3], s32 offset:128 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v189, off, s[0:3], s32 offset:132 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v188, off, s[0:3], s32 offset:136 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v187, off, s[0:3], s32 offset:140 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v186, off, s[0:3], s32 offset:144 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v185, off, s[0:3], s32 offset:148 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v184, off, s[0:3], s32 offset:152 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v175, off, s[0:3], s32 offset:156 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v174, off, s[0:3], s32 offset:160 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v173, off, s[0:3], s32 offset:164 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v172, off, s[0:3], s32 offset:168 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v171, off, s[0:3], s32 offset:172 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v170, off, s[0:3], s32 offset:176 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v169, off, s[0:3], s32 offset:180 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v168, off, s[0:3], s32 offset:184 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v159, off, s[0:3], s32 offset:188 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v158, off, s[0:3], s32 offset:192 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v157, off, s[0:3], s32 offset:196 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v156, off, s[0:3], s32 offset:200 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v155, off, s[0:3], s32 offset:204 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v154, off, s[0:3], s32 offset:208 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v153, off, s[0:3], s32 offset:212 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v152, off, s[0:3], s32 offset:216 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v143, off, s[0:3], s32 offset:220 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v142, off, s[0:3], s32 offset:224 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v141, off, s[0:3], s32 offset:228 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v140, off, s[0:3], s32 offset:232 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v139, off, s[0:3], s32 offset:236 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v138, off, s[0:3], s32 offset:240 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v137, off, s[0:3], s32 offset:244 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v136, off, s[0:3], s32 offset:248 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v127, off, s[0:3], s32 offset:252 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v126, off, s[0:3], s32 offset:256 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v125, off, s[0:3], s32 offset:260 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v124, off, s[0:3], s32 offset:264 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v123, off, s[0:3], s32 offset:268 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v122, off, s[0:3], s32 offset:272 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v121, off, s[0:3], s32 offset:276 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v120, off, s[0:3], s32 offset:280 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v111, off, s[0:3], s32 offset:284 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v110, off, s[0:3], s32 offset:288 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v109, off, s[0:3], s32 offset:292 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v108, off, s[0:3], s32 offset:296 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v107, off, s[0:3], s32 offset:300 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v106, off, s[0:3], s32 offset:304 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v105, off, s[0:3], s32 offset:308 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v104, off, s[0:3], s32 offset:312 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v95, off, s[0:3], s32 offset:316 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v94, off, s[0:3], s32 offset:320 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v93, off, s[0:3], s32 offset:324 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v92, off, s[0:3], s32 offset:328 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v91, off, s[0:3], s32 offset:332 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v90, off, s[0:3], s32 offset:336 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v89, off, s[0:3], s32 offset:340 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v88, off, s[0:3], s32 offset:344 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v79, off, s[0:3], s32 offset:348 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v78, off, s[0:3], s32 offset:352 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v77, off, s[0:3], s32 offset:356 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v76, off, s[0:3], s32 offset:360 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v75, off, s[0:3], s32 offset:364 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v74, off, s[0:3], s32 offset:368 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v73, off, s[0:3], s32 offset:372 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v72, off, s[0:3], s32 offset:376 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v63, off, s[0:3], s32 offset:380 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v62, off, s[0:3], s32 offset:384 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v61, off, s[0:3], s32 offset:388 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v60, off, s[0:3], s32 offset:392 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v59, off, s[0:3], s32 offset:396 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v58, off, s[0:3], s32 offset:400 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v57, off, s[0:3], s32 offset:404 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v56, off, s[0:3], s32 offset:408 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v47, off, s[0:3], s32 offset:412 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v46, off, s[0:3], s32 offset:416 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v45, off, s[0:3], s32 offset:420 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v44, off, s[0:3], s32 offset:424 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v43, off, s[0:3], s32 offset:428 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v42, off, s[0:3], s32 offset:432 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v41, off, s[0:3], s32 offset:436 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v40, off, s[0:3], s32 offset:440 ; 4-byte Folded Reload
+; GCN-NEXT: s_waitcnt lgkmcnt(0)
+; GCN-NEXT: s_setpc_b64 s[4:5]
%alloca = alloca i32, align 4, addrspace(5)
store volatile i32 0, i32 addrspace(5)* %alloca
ret void
}
-; Special case where all registers are explicitly clobbered in the function and
-; we have no VGPR to allocate for SGPR spills. We are forced to spill to memory.
-
-; GCN-LABEL: {{^}}spill_sgpr_no_free_vgpr:
-; GCN: v_writelane_b32 v{{[0-9]+}}, s34, 0
-; GCN: v_writelane_b32 v{{[0-9]+}}, s35, 1
-; GCN: v_writelane_b32 v{{[0-9]+}}, s36, 2
-; GCN: v_writelane_b32 v{{[0-9]+}}, s37, 3
-; GCN: buffer_store_dword v{{[0-9]+}}, off, s[0:3], s32
-; GCN: buffer_store_dword v{{[0-9]+}}, off, s[0:3], s32
-; GCN: buffer_store_dword v{{[0-9]+}}, off, s[0:3], s32
-; GCN: buffer_store_dword v{{[0-9]+}}, off, s[0:3], s32
-; GCN: #ASMEND
-; GCN: buffer_load_dword v{{[0-9]+}}
-; GCN: buffer_load_dword v{{[0-9]+}}
-; GCN: buffer_load_dword v{{[0-9]+}}
-; GCN: buffer_load_dword v{{[0-9]+}}
-; GCN: v_readlane_b32 s37, v{{[0-9]+}}, 3
-; GCN: v_readlane_b32 s36, v{{[0-9]+}}, 2
-; GCN: v_readlane_b32 s35, v{{[0-9]+}}, 1
-; GCN: v_readlane_b32 s34, v{{[0-9]+}}, 0
-
define void @spill_sgpr_no_free_vgpr(<4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %in) #0 {
+; GCN-LABEL: spill_sgpr_no_free_vgpr:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: s_or_saveexec_b64 s[4:5], -1
+; GCN-NEXT: buffer_store_dword v4, off, s[0:3], s32 offset:448 ; 4-byte Folded Spill
+; GCN-NEXT: s_mov_b64 exec, s[4:5]
+; GCN-NEXT: buffer_store_dword v40, off, s[0:3], s32 offset:444 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v41, off, s[0:3], s32 offset:440 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v42, off, s[0:3], s32 offset:436 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v43, off, s[0:3], s32 offset:432 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v44, off, s[0:3], s32 offset:428 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v45, off, s[0:3], s32 offset:424 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v46, off, s[0:3], s32 offset:420 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v47, off, s[0:3], s32 offset:416 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v56, off, s[0:3], s32 offset:412 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v57, off, s[0:3], s32 offset:408 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v58, off, s[0:3], s32 offset:404 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v59, off, s[0:3], s32 offset:400 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v60, off, s[0:3], s32 offset:396 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v61, off, s[0:3], s32 offset:392 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v62, off, s[0:3], s32 offset:388 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v63, off, s[0:3], s32 offset:384 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v72, off, s[0:3], s32 offset:380 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v73, off, s[0:3], s32 offset:376 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v74, off, s[0:3], s32 offset:372 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v75, off, s[0:3], s32 offset:368 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v76, off, s[0:3], s32 offset:364 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v77, off, s[0:3], s32 offset:360 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v78, off, s[0:3], s32 offset:356 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v79, off, s[0:3], s32 offset:352 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v88, off, s[0:3], s32 offset:348 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v89, off, s[0:3], s32 offset:344 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v90, off, s[0:3], s32 offset:340 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v91, off, s[0:3], s32 offset:336 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v92, off, s[0:3], s32 offset:332 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v93, off, s[0:3], s32 offset:328 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v94, off, s[0:3], s32 offset:324 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v95, off, s[0:3], s32 offset:320 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v104, off, s[0:3], s32 offset:316 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v105, off, s[0:3], s32 offset:312 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v106, off, s[0:3], s32 offset:308 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v107, off, s[0:3], s32 offset:304 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v108, off, s[0:3], s32 offset:300 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v109, off, s[0:3], s32 offset:296 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v110, off, s[0:3], s32 offset:292 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v111, off, s[0:3], s32 offset:288 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v120, off, s[0:3], s32 offset:284 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v121, off, s[0:3], s32 offset:280 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v122, off, s[0:3], s32 offset:276 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v123, off, s[0:3], s32 offset:272 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v124, off, s[0:3], s32 offset:268 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v125, off, s[0:3], s32 offset:264 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v126, off, s[0:3], s32 offset:260 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v127, off, s[0:3], s32 offset:256 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v136, off, s[0:3], s32 offset:252 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v137, off, s[0:3], s32 offset:248 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v138, off, s[0:3], s32 offset:244 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v139, off, s[0:3], s32 offset:240 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v140, off, s[0:3], s32 offset:236 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v141, off, s[0:3], s32 offset:232 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v142, off, s[0:3], s32 offset:228 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v143, off, s[0:3], s32 offset:224 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v152, off, s[0:3], s32 offset:220 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v153, off, s[0:3], s32 offset:216 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v154, off, s[0:3], s32 offset:212 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v155, off, s[0:3], s32 offset:208 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v156, off, s[0:3], s32 offset:204 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v157, off, s[0:3], s32 offset:200 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v158, off, s[0:3], s32 offset:196 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v159, off, s[0:3], s32 offset:192 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v168, off, s[0:3], s32 offset:188 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v169, off, s[0:3], s32 offset:184 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v170, off, s[0:3], s32 offset:180 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v171, off, s[0:3], s32 offset:176 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v172, off, s[0:3], s32 offset:172 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v173, off, s[0:3], s32 offset:168 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v174, off, s[0:3], s32 offset:164 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v175, off, s[0:3], s32 offset:160 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v184, off, s[0:3], s32 offset:156 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v185, off, s[0:3], s32 offset:152 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v186, off, s[0:3], s32 offset:148 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v187, off, s[0:3], s32 offset:144 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v188, off, s[0:3], s32 offset:140 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v189, off, s[0:3], s32 offset:136 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v190, off, s[0:3], s32 offset:132 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v191, off, s[0:3], s32 offset:128 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v200, off, s[0:3], s32 offset:124 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v201, off, s[0:3], s32 offset:120 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v202, off, s[0:3], s32 offset:116 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v203, off, s[0:3], s32 offset:112 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v204, off, s[0:3], s32 offset:108 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v205, off, s[0:3], s32 offset:104 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v206, off, s[0:3], s32 offset:100 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v207, off, s[0:3], s32 offset:96 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v216, off, s[0:3], s32 offset:92 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v217, off, s[0:3], s32 offset:88 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v218, off, s[0:3], s32 offset:84 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v219, off, s[0:3], s32 offset:80 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v220, off, s[0:3], s32 offset:76 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v221, off, s[0:3], s32 offset:72 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v222, off, s[0:3], s32 offset:68 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v223, off, s[0:3], s32 offset:64 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v232, off, s[0:3], s32 offset:60 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v233, off, s[0:3], s32 offset:56 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v234, off, s[0:3], s32 offset:52 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v235, off, s[0:3], s32 offset:48 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v236, off, s[0:3], s32 offset:44 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v237, off, s[0:3], s32 offset:40 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v238, off, s[0:3], s32 offset:36 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v239, off, s[0:3], s32 offset:32 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v248, off, s[0:3], s32 offset:28 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v249, off, s[0:3], s32 offset:24 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v250, off, s[0:3], s32 offset:20 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v251, off, s[0:3], s32 offset:16 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v252, off, s[0:3], s32 offset:12 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v253, off, s[0:3], s32 offset:8 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v254, off, s[0:3], s32 offset:4 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v255, off, s[0:3], s32 ; 4-byte Folded Spill
+; GCN-NEXT: v_writelane_b32 v4, s34, 0
+; GCN-NEXT: v_writelane_b32 v4, s35, 1
+; GCN-NEXT: v_writelane_b32 v4, s36, 2
+; GCN-NEXT: v_writelane_b32 v4, s37, 3
+; GCN-NEXT: v_mov_b32_e32 v5, v3
+; GCN-NEXT: v_mov_b32_e32 v3, v1
+; GCN-NEXT: ; implicit-def: $sgpr4
+; GCN-NEXT: ; implicit-def: $sgpr4
+; GCN-NEXT: ; kill: def $vgpr3 killed $vgpr3 killed $exec
+; GCN-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec
+; GCN-NEXT: v_mov_b32_e32 v1, v3
+; GCN-NEXT: ; implicit-def: $sgpr4
+; GCN-NEXT: ; implicit-def: $sgpr4
+; GCN-NEXT: ; kill: def $vgpr5 killed $vgpr5 killed $exec
+; GCN-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec
+; GCN-NEXT: v_mov_b32_e32 v3, v5
+; GCN-NEXT: ; implicit-def: $sgpr4_sgpr5
+; GCN-NEXT: ; implicit-def: $sgpr4_sgpr5
+; GCN-NEXT: flat_load_dwordx4 v[5:8], v[2:3]
+; GCN-NEXT: s_waitcnt vmcnt(0)
+; GCN-NEXT: buffer_store_dword v5, off, s[0:3], s32 offset:452 ; 4-byte Folded Spill
+; GCN-NEXT: s_waitcnt vmcnt(0)
+; GCN-NEXT: buffer_store_dword v6, off, s[0:3], s32 offset:456 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v7, off, s[0:3], s32 offset:460 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v8, off, s[0:3], s32 offset:464 ; 4-byte Folded Spill
+; GCN-NEXT: ;;#ASMSTART
+; GCN-NEXT: ;;#ASMEND
+; GCN-NEXT: buffer_load_dword v5, off, s[0:3], s32 offset:452 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v6, off, s[0:3], s32 offset:456 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v7, off, s[0:3], s32 offset:460 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v8, off, s[0:3], s32 offset:464 ; 4-byte Folded Reload
+; GCN-NEXT: ;;#ASMSTART
+; GCN-NEXT: ;;#ASMEND
+; GCN-NEXT: s_waitcnt vmcnt(0)
+; GCN-NEXT: flat_store_dwordx4 v[0:1], v[5:8]
+; GCN-NEXT: v_readlane_b32 s37, v4, 3
+; GCN-NEXT: v_readlane_b32 s36, v4, 2
+; GCN-NEXT: v_readlane_b32 s35, v4, 1
+; GCN-NEXT: v_readlane_b32 s34, v4, 0
+; GCN-NEXT: buffer_load_dword v255, off, s[0:3], s32 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v254, off, s[0:3], s32 offset:4 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v253, off, s[0:3], s32 offset:8 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v252, off, s[0:3], s32 offset:12 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v251, off, s[0:3], s32 offset:16 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v250, off, s[0:3], s32 offset:20 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v249, off, s[0:3], s32 offset:24 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v248, off, s[0:3], s32 offset:28 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v239, off, s[0:3], s32 offset:32 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v238, off, s[0:3], s32 offset:36 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v237, off, s[0:3], s32 offset:40 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v236, off, s[0:3], s32 offset:44 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v235, off, s[0:3], s32 offset:48 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v234, off, s[0:3], s32 offset:52 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v233, off, s[0:3], s32 offset:56 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v232, off, s[0:3], s32 offset:60 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v223, off, s[0:3], s32 offset:64 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v222, off, s[0:3], s32 offset:68 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v221, off, s[0:3], s32 offset:72 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v220, off, s[0:3], s32 offset:76 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v219, off, s[0:3], s32 offset:80 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v218, off, s[0:3], s32 offset:84 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v217, off, s[0:3], s32 offset:88 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v216, off, s[0:3], s32 offset:92 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v207, off, s[0:3], s32 offset:96 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v206, off, s[0:3], s32 offset:100 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v205, off, s[0:3], s32 offset:104 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v204, off, s[0:3], s32 offset:108 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v203, off, s[0:3], s32 offset:112 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v202, off, s[0:3], s32 offset:116 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v201, off, s[0:3], s32 offset:120 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v200, off, s[0:3], s32 offset:124 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v191, off, s[0:3], s32 offset:128 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v190, off, s[0:3], s32 offset:132 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v189, off, s[0:3], s32 offset:136 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v188, off, s[0:3], s32 offset:140 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v187, off, s[0:3], s32 offset:144 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v186, off, s[0:3], s32 offset:148 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v185, off, s[0:3], s32 offset:152 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v184, off, s[0:3], s32 offset:156 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v175, off, s[0:3], s32 offset:160 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v174, off, s[0:3], s32 offset:164 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v173, off, s[0:3], s32 offset:168 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v172, off, s[0:3], s32 offset:172 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v171, off, s[0:3], s32 offset:176 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v170, off, s[0:3], s32 offset:180 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v169, off, s[0:3], s32 offset:184 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v168, off, s[0:3], s32 offset:188 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v159, off, s[0:3], s32 offset:192 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v158, off, s[0:3], s32 offset:196 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v157, off, s[0:3], s32 offset:200 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v156, off, s[0:3], s32 offset:204 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v155, off, s[0:3], s32 offset:208 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v154, off, s[0:3], s32 offset:212 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v153, off, s[0:3], s32 offset:216 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v152, off, s[0:3], s32 offset:220 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v143, off, s[0:3], s32 offset:224 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v142, off, s[0:3], s32 offset:228 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v141, off, s[0:3], s32 offset:232 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v140, off, s[0:3], s32 offset:236 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v139, off, s[0:3], s32 offset:240 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v138, off, s[0:3], s32 offset:244 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v137, off, s[0:3], s32 offset:248 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v136, off, s[0:3], s32 offset:252 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v127, off, s[0:3], s32 offset:256 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v126, off, s[0:3], s32 offset:260 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v125, off, s[0:3], s32 offset:264 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v124, off, s[0:3], s32 offset:268 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v123, off, s[0:3], s32 offset:272 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v122, off, s[0:3], s32 offset:276 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v121, off, s[0:3], s32 offset:280 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v120, off, s[0:3], s32 offset:284 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v111, off, s[0:3], s32 offset:288 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v110, off, s[0:3], s32 offset:292 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v109, off, s[0:3], s32 offset:296 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v108, off, s[0:3], s32 offset:300 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v107, off, s[0:3], s32 offset:304 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v106, off, s[0:3], s32 offset:308 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v105, off, s[0:3], s32 offset:312 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v104, off, s[0:3], s32 offset:316 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v95, off, s[0:3], s32 offset:320 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v94, off, s[0:3], s32 offset:324 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v93, off, s[0:3], s32 offset:328 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v92, off, s[0:3], s32 offset:332 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v91, off, s[0:3], s32 offset:336 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v90, off, s[0:3], s32 offset:340 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v89, off, s[0:3], s32 offset:344 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v88, off, s[0:3], s32 offset:348 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v79, off, s[0:3], s32 offset:352 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v78, off, s[0:3], s32 offset:356 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v77, off, s[0:3], s32 offset:360 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v76, off, s[0:3], s32 offset:364 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v75, off, s[0:3], s32 offset:368 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v74, off, s[0:3], s32 offset:372 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v73, off, s[0:3], s32 offset:376 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v72, off, s[0:3], s32 offset:380 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v63, off, s[0:3], s32 offset:384 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v62, off, s[0:3], s32 offset:388 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v61, off, s[0:3], s32 offset:392 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v60, off, s[0:3], s32 offset:396 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v59, off, s[0:3], s32 offset:400 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v58, off, s[0:3], s32 offset:404 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v57, off, s[0:3], s32 offset:408 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v56, off, s[0:3], s32 offset:412 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v47, off, s[0:3], s32 offset:416 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v46, off, s[0:3], s32 offset:420 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v45, off, s[0:3], s32 offset:424 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v44, off, s[0:3], s32 offset:428 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v43, off, s[0:3], s32 offset:432 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v42, off, s[0:3], s32 offset:436 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v41, off, s[0:3], s32 offset:440 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v40, off, s[0:3], s32 offset:444 ; 4-byte Folded Reload
+; GCN-NEXT: s_or_saveexec_b64 s[4:5], -1
+; GCN-NEXT: buffer_load_dword v4, off, s[0:3], s32 offset:448 ; 4-byte Folded Reload
+; GCN-NEXT: s_mov_b64 exec, s[4:5]
+; GCN-NEXT: s_waitcnt vmcnt(0)
+; GCN-NEXT: s_setpc_b64 s[30:31]
%a = load <4 x i32>, <4 x i32> addrspace(1)* %in
call void asm sideeffect "",
"~{v6},~{v7},~{v8},~{v9}
ret void
}
-; If IPRA no-CSR optimization is enabled, we will not be able to allocate an
-; SGPR for VGPR spills in the parent function since this child function uses all
-; VGPRs.
-
define internal void @child_function_ipra() #0 {
+; GCN-LABEL: child_function_ipra:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: ;;#ASMSTART
+; GCN-NEXT: ;;#ASMEND
+; GCN-NEXT: s_setpc_b64 s[30:31]
call void asm sideeffect "",
"~{v0},~{v1},~{v2},~{v3},~{v4},~{v5},~{v6},~{v7},~{v8},~{v9}
,~{v10},~{v11},~{v12},~{v13},~{v14},~{v15},~{v16},~{v17},~{v18},~{v19}
ret void
}
-; GCN-LABEL: {{^}}spill_sgpr_no_free_vgpr_ipra:
-; GCN: v_writelane_b32 v0, s30, 0
-; GCN: buffer_store_dword v0, off
-; GCN: v_writelane_b32 v0, s31, 0
-; GCN: buffer_store_dword v0, off
-; GCN: swappc
-; GCN: buffer_load_dword v0, off
-; GCN: v_readlane_b32 s31, v0, 0
-; GCN: buffer_load_dword v0, off
-; GCN: v_readlane_b32 s30, v0, 0
define void @spill_sgpr_no_free_vgpr_ipra() #0 {
+; GCN-LABEL: spill_sgpr_no_free_vgpr_ipra:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: s_mov_b32 s6, s33
+; GCN-NEXT: s_mov_b32 s33, s32
+; GCN-NEXT: s_add_i32 s32, s32, 0x7400
+; GCN-NEXT: buffer_store_dword v40, off, s[0:3], s33 offset:444 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v41, off, s[0:3], s33 offset:440 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v42, off, s[0:3], s33 offset:436 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v43, off, s[0:3], s33 offset:432 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v44, off, s[0:3], s33 offset:428 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v45, off, s[0:3], s33 offset:424 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v46, off, s[0:3], s33 offset:420 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v47, off, s[0:3], s33 offset:416 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v56, off, s[0:3], s33 offset:412 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v57, off, s[0:3], s33 offset:408 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v58, off, s[0:3], s33 offset:404 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v59, off, s[0:3], s33 offset:400 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v60, off, s[0:3], s33 offset:396 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v61, off, s[0:3], s33 offset:392 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v62, off, s[0:3], s33 offset:388 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v63, off, s[0:3], s33 offset:384 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v72, off, s[0:3], s33 offset:380 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v73, off, s[0:3], s33 offset:376 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v74, off, s[0:3], s33 offset:372 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v75, off, s[0:3], s33 offset:368 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v76, off, s[0:3], s33 offset:364 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v77, off, s[0:3], s33 offset:360 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v78, off, s[0:3], s33 offset:356 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v79, off, s[0:3], s33 offset:352 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v88, off, s[0:3], s33 offset:348 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v89, off, s[0:3], s33 offset:344 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v90, off, s[0:3], s33 offset:340 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v91, off, s[0:3], s33 offset:336 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v92, off, s[0:3], s33 offset:332 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v93, off, s[0:3], s33 offset:328 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v94, off, s[0:3], s33 offset:324 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v95, off, s[0:3], s33 offset:320 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v104, off, s[0:3], s33 offset:316 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v105, off, s[0:3], s33 offset:312 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v106, off, s[0:3], s33 offset:308 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v107, off, s[0:3], s33 offset:304 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v108, off, s[0:3], s33 offset:300 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v109, off, s[0:3], s33 offset:296 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v110, off, s[0:3], s33 offset:292 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v111, off, s[0:3], s33 offset:288 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v120, off, s[0:3], s33 offset:284 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v121, off, s[0:3], s33 offset:280 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v122, off, s[0:3], s33 offset:276 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v123, off, s[0:3], s33 offset:272 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v124, off, s[0:3], s33 offset:268 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v125, off, s[0:3], s33 offset:264 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v126, off, s[0:3], s33 offset:260 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v127, off, s[0:3], s33 offset:256 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v136, off, s[0:3], s33 offset:252 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v137, off, s[0:3], s33 offset:248 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v138, off, s[0:3], s33 offset:244 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v139, off, s[0:3], s33 offset:240 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v140, off, s[0:3], s33 offset:236 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v141, off, s[0:3], s33 offset:232 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v142, off, s[0:3], s33 offset:228 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v143, off, s[0:3], s33 offset:224 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v152, off, s[0:3], s33 offset:220 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v153, off, s[0:3], s33 offset:216 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v154, off, s[0:3], s33 offset:212 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v155, off, s[0:3], s33 offset:208 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v156, off, s[0:3], s33 offset:204 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v157, off, s[0:3], s33 offset:200 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v158, off, s[0:3], s33 offset:196 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v159, off, s[0:3], s33 offset:192 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v168, off, s[0:3], s33 offset:188 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v169, off, s[0:3], s33 offset:184 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v170, off, s[0:3], s33 offset:180 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v171, off, s[0:3], s33 offset:176 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v172, off, s[0:3], s33 offset:172 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v173, off, s[0:3], s33 offset:168 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v174, off, s[0:3], s33 offset:164 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v175, off, s[0:3], s33 offset:160 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v184, off, s[0:3], s33 offset:156 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v185, off, s[0:3], s33 offset:152 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v186, off, s[0:3], s33 offset:148 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v187, off, s[0:3], s33 offset:144 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v188, off, s[0:3], s33 offset:140 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v189, off, s[0:3], s33 offset:136 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v190, off, s[0:3], s33 offset:132 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v191, off, s[0:3], s33 offset:128 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v200, off, s[0:3], s33 offset:124 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v201, off, s[0:3], s33 offset:120 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v202, off, s[0:3], s33 offset:116 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v203, off, s[0:3], s33 offset:112 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v204, off, s[0:3], s33 offset:108 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v205, off, s[0:3], s33 offset:104 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v206, off, s[0:3], s33 offset:100 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v207, off, s[0:3], s33 offset:96 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v216, off, s[0:3], s33 offset:92 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v217, off, s[0:3], s33 offset:88 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v218, off, s[0:3], s33 offset:84 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v219, off, s[0:3], s33 offset:80 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v220, off, s[0:3], s33 offset:76 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v221, off, s[0:3], s33 offset:72 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v222, off, s[0:3], s33 offset:68 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v223, off, s[0:3], s33 offset:64 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v232, off, s[0:3], s33 offset:60 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v233, off, s[0:3], s33 offset:56 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v234, off, s[0:3], s33 offset:52 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v235, off, s[0:3], s33 offset:48 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v236, off, s[0:3], s33 offset:44 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v237, off, s[0:3], s33 offset:40 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v238, off, s[0:3], s33 offset:36 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v239, off, s[0:3], s33 offset:32 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v248, off, s[0:3], s33 offset:28 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v249, off, s[0:3], s33 offset:24 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v250, off, s[0:3], s33 offset:20 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v251, off, s[0:3], s33 offset:16 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v252, off, s[0:3], s33 offset:12 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v253, off, s[0:3], s33 offset:8 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v254, off, s[0:3], s33 offset:4 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v255, off, s[0:3], s33 ; 4-byte Folded Spill
+; GCN-NEXT: s_mov_b64 s[14:15], exec
+; GCN-NEXT: s_mov_b64 exec, 1
+; GCN-NEXT: buffer_store_dword v1, off, s[0:3], s33 offset:456
+; GCN-NEXT: v_writelane_b32 v1, s30, 0
+; GCN-NEXT: buffer_store_dword v1, off, s[0:3], s33 offset:448 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_load_dword v1, off, s[0:3], s33 offset:456
+; GCN-NEXT: s_waitcnt vmcnt(0)
+; GCN-NEXT: s_mov_b64 exec, s[14:15]
+; GCN-NEXT: s_mov_b64 s[12:13], exec
+; GCN-NEXT: s_mov_b64 exec, 1
+; GCN-NEXT: buffer_store_dword v0, off, s[0:3], s33 offset:456
+; GCN-NEXT: v_writelane_b32 v0, s31, 0
+; GCN-NEXT: buffer_store_dword v0, off, s[0:3], s33 offset:452 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_load_dword v0, off, s[0:3], s33 offset:456
+; GCN-NEXT: s_waitcnt vmcnt(0)
+; GCN-NEXT: s_mov_b64 exec, s[12:13]
+; GCN-NEXT: s_getpc_b64 s[4:5]
+; GCN-NEXT: s_add_u32 s4, s4, child_function_ipra@rel32@lo+4
+; GCN-NEXT: s_addc_u32 s5, s5, child_function_ipra@rel32@hi+12
+; GCN-NEXT: s_mov_b64 s[10:11], s[2:3]
+; GCN-NEXT: s_mov_b64 s[8:9], s[0:1]
+; GCN-NEXT: s_mov_b64 s[0:1], s[8:9]
+; GCN-NEXT: s_mov_b64 s[2:3], s[10:11]
+; GCN-NEXT: s_swappc_b64 s[30:31], s[4:5]
+; GCN-NEXT: s_mov_b64 s[8:9], exec
+; GCN-NEXT: s_mov_b64 exec, 1
+; GCN-NEXT: buffer_store_dword v1, off, s[0:3], s33 offset:456
+; GCN-NEXT: buffer_load_dword v1, off, s[0:3], s33 offset:452 ; 4-byte Folded Reload
+; GCN-NEXT: s_waitcnt vmcnt(0)
+; GCN-NEXT: v_readlane_b32 s31, v1, 0
+; GCN-NEXT: buffer_load_dword v1, off, s[0:3], s33 offset:456
+; GCN-NEXT: s_waitcnt vmcnt(0)
+; GCN-NEXT: s_mov_b64 exec, s[8:9]
+; GCN-NEXT: s_mov_b64 s[4:5], exec
+; GCN-NEXT: s_mov_b64 exec, 1
+; GCN-NEXT: buffer_store_dword v0, off, s[0:3], s33 offset:456
+; GCN-NEXT: buffer_load_dword v0, off, s[0:3], s33 offset:448 ; 4-byte Folded Reload
+; GCN-NEXT: s_waitcnt vmcnt(0)
+; GCN-NEXT: v_readlane_b32 s30, v0, 0
+; GCN-NEXT: buffer_load_dword v0, off, s[0:3], s33 offset:456
+; GCN-NEXT: s_waitcnt vmcnt(0)
+; GCN-NEXT: s_mov_b64 exec, s[4:5]
+; GCN-NEXT: buffer_load_dword v255, off, s[0:3], s33 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v254, off, s[0:3], s33 offset:4 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v253, off, s[0:3], s33 offset:8 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v252, off, s[0:3], s33 offset:12 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v251, off, s[0:3], s33 offset:16 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v250, off, s[0:3], s33 offset:20 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v249, off, s[0:3], s33 offset:24 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v248, off, s[0:3], s33 offset:28 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v239, off, s[0:3], s33 offset:32 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v238, off, s[0:3], s33 offset:36 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v237, off, s[0:3], s33 offset:40 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v236, off, s[0:3], s33 offset:44 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v235, off, s[0:3], s33 offset:48 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v234, off, s[0:3], s33 offset:52 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v233, off, s[0:3], s33 offset:56 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v232, off, s[0:3], s33 offset:60 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v223, off, s[0:3], s33 offset:64 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v222, off, s[0:3], s33 offset:68 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v221, off, s[0:3], s33 offset:72 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v220, off, s[0:3], s33 offset:76 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v219, off, s[0:3], s33 offset:80 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v218, off, s[0:3], s33 offset:84 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v217, off, s[0:3], s33 offset:88 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v216, off, s[0:3], s33 offset:92 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v207, off, s[0:3], s33 offset:96 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v206, off, s[0:3], s33 offset:100 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v205, off, s[0:3], s33 offset:104 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v204, off, s[0:3], s33 offset:108 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v203, off, s[0:3], s33 offset:112 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v202, off, s[0:3], s33 offset:116 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v201, off, s[0:3], s33 offset:120 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v200, off, s[0:3], s33 offset:124 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v191, off, s[0:3], s33 offset:128 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v190, off, s[0:3], s33 offset:132 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v189, off, s[0:3], s33 offset:136 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v188, off, s[0:3], s33 offset:140 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v187, off, s[0:3], s33 offset:144 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v186, off, s[0:3], s33 offset:148 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v185, off, s[0:3], s33 offset:152 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v184, off, s[0:3], s33 offset:156 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v175, off, s[0:3], s33 offset:160 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v174, off, s[0:3], s33 offset:164 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v173, off, s[0:3], s33 offset:168 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v172, off, s[0:3], s33 offset:172 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v171, off, s[0:3], s33 offset:176 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v170, off, s[0:3], s33 offset:180 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v169, off, s[0:3], s33 offset:184 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v168, off, s[0:3], s33 offset:188 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v159, off, s[0:3], s33 offset:192 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v158, off, s[0:3], s33 offset:196 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v157, off, s[0:3], s33 offset:200 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v156, off, s[0:3], s33 offset:204 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v155, off, s[0:3], s33 offset:208 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v154, off, s[0:3], s33 offset:212 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v153, off, s[0:3], s33 offset:216 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v152, off, s[0:3], s33 offset:220 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v143, off, s[0:3], s33 offset:224 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v142, off, s[0:3], s33 offset:228 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v141, off, s[0:3], s33 offset:232 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v140, off, s[0:3], s33 offset:236 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v139, off, s[0:3], s33 offset:240 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v138, off, s[0:3], s33 offset:244 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v137, off, s[0:3], s33 offset:248 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v136, off, s[0:3], s33 offset:252 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v127, off, s[0:3], s33 offset:256 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v126, off, s[0:3], s33 offset:260 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v125, off, s[0:3], s33 offset:264 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v124, off, s[0:3], s33 offset:268 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v123, off, s[0:3], s33 offset:272 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v122, off, s[0:3], s33 offset:276 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v121, off, s[0:3], s33 offset:280 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v120, off, s[0:3], s33 offset:284 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v111, off, s[0:3], s33 offset:288 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v110, off, s[0:3], s33 offset:292 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v109, off, s[0:3], s33 offset:296 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v108, off, s[0:3], s33 offset:300 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v107, off, s[0:3], s33 offset:304 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v106, off, s[0:3], s33 offset:308 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v105, off, s[0:3], s33 offset:312 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v104, off, s[0:3], s33 offset:316 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v95, off, s[0:3], s33 offset:320 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v94, off, s[0:3], s33 offset:324 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v93, off, s[0:3], s33 offset:328 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v92, off, s[0:3], s33 offset:332 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v91, off, s[0:3], s33 offset:336 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v90, off, s[0:3], s33 offset:340 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v89, off, s[0:3], s33 offset:344 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v88, off, s[0:3], s33 offset:348 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v79, off, s[0:3], s33 offset:352 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v78, off, s[0:3], s33 offset:356 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v77, off, s[0:3], s33 offset:360 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v76, off, s[0:3], s33 offset:364 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v75, off, s[0:3], s33 offset:368 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v74, off, s[0:3], s33 offset:372 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v73, off, s[0:3], s33 offset:376 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v72, off, s[0:3], s33 offset:380 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v63, off, s[0:3], s33 offset:384 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v62, off, s[0:3], s33 offset:388 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v61, off, s[0:3], s33 offset:392 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v60, off, s[0:3], s33 offset:396 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v59, off, s[0:3], s33 offset:400 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v58, off, s[0:3], s33 offset:404 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v57, off, s[0:3], s33 offset:408 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v56, off, s[0:3], s33 offset:412 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v47, off, s[0:3], s33 offset:416 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v46, off, s[0:3], s33 offset:420 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v45, off, s[0:3], s33 offset:424 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v44, off, s[0:3], s33 offset:428 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v43, off, s[0:3], s33 offset:432 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v42, off, s[0:3], s33 offset:436 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v41, off, s[0:3], s33 offset:440 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v40, off, s[0:3], s33 offset:444 ; 4-byte Folded Reload
+; GCN-NEXT: s_add_i32 s32, s32, 0xffff8c00
+; GCN-NEXT: s_mov_b32 s33, s6
+; GCN-NEXT: s_waitcnt vmcnt(0)
+; GCN-NEXT: s_setpc_b64 s[30:31]
call void @child_function_ipra()
ret void
}
define internal void @child_function_ipra_tail_call() #0 {
+; GCN-LABEL: child_function_ipra_tail_call:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: buffer_store_dword v40, off, s[0:3], s32 offset:444 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v41, off, s[0:3], s32 offset:440 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v42, off, s[0:3], s32 offset:436 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v43, off, s[0:3], s32 offset:432 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v44, off, s[0:3], s32 offset:428 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v45, off, s[0:3], s32 offset:424 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v46, off, s[0:3], s32 offset:420 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v47, off, s[0:3], s32 offset:416 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v56, off, s[0:3], s32 offset:412 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v57, off, s[0:3], s32 offset:408 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v58, off, s[0:3], s32 offset:404 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v59, off, s[0:3], s32 offset:400 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v60, off, s[0:3], s32 offset:396 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v61, off, s[0:3], s32 offset:392 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v62, off, s[0:3], s32 offset:388 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v63, off, s[0:3], s32 offset:384 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v72, off, s[0:3], s32 offset:380 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v73, off, s[0:3], s32 offset:376 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v74, off, s[0:3], s32 offset:372 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v75, off, s[0:3], s32 offset:368 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v76, off, s[0:3], s32 offset:364 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v77, off, s[0:3], s32 offset:360 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v78, off, s[0:3], s32 offset:356 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v79, off, s[0:3], s32 offset:352 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v88, off, s[0:3], s32 offset:348 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v89, off, s[0:3], s32 offset:344 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v90, off, s[0:3], s32 offset:340 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v91, off, s[0:3], s32 offset:336 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v92, off, s[0:3], s32 offset:332 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v93, off, s[0:3], s32 offset:328 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v94, off, s[0:3], s32 offset:324 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v95, off, s[0:3], s32 offset:320 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v104, off, s[0:3], s32 offset:316 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v105, off, s[0:3], s32 offset:312 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v106, off, s[0:3], s32 offset:308 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v107, off, s[0:3], s32 offset:304 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v108, off, s[0:3], s32 offset:300 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v109, off, s[0:3], s32 offset:296 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v110, off, s[0:3], s32 offset:292 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v111, off, s[0:3], s32 offset:288 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v120, off, s[0:3], s32 offset:284 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v121, off, s[0:3], s32 offset:280 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v122, off, s[0:3], s32 offset:276 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v123, off, s[0:3], s32 offset:272 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v124, off, s[0:3], s32 offset:268 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v125, off, s[0:3], s32 offset:264 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v126, off, s[0:3], s32 offset:260 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v127, off, s[0:3], s32 offset:256 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v136, off, s[0:3], s32 offset:252 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v137, off, s[0:3], s32 offset:248 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v138, off, s[0:3], s32 offset:244 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v139, off, s[0:3], s32 offset:240 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v140, off, s[0:3], s32 offset:236 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v141, off, s[0:3], s32 offset:232 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v142, off, s[0:3], s32 offset:228 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v143, off, s[0:3], s32 offset:224 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v152, off, s[0:3], s32 offset:220 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v153, off, s[0:3], s32 offset:216 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v154, off, s[0:3], s32 offset:212 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v155, off, s[0:3], s32 offset:208 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v156, off, s[0:3], s32 offset:204 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v157, off, s[0:3], s32 offset:200 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v158, off, s[0:3], s32 offset:196 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v159, off, s[0:3], s32 offset:192 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v168, off, s[0:3], s32 offset:188 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v169, off, s[0:3], s32 offset:184 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v170, off, s[0:3], s32 offset:180 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v171, off, s[0:3], s32 offset:176 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v172, off, s[0:3], s32 offset:172 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v173, off, s[0:3], s32 offset:168 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v174, off, s[0:3], s32 offset:164 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v175, off, s[0:3], s32 offset:160 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v184, off, s[0:3], s32 offset:156 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v185, off, s[0:3], s32 offset:152 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v186, off, s[0:3], s32 offset:148 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v187, off, s[0:3], s32 offset:144 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v188, off, s[0:3], s32 offset:140 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v189, off, s[0:3], s32 offset:136 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v190, off, s[0:3], s32 offset:132 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v191, off, s[0:3], s32 offset:128 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v200, off, s[0:3], s32 offset:124 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v201, off, s[0:3], s32 offset:120 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v202, off, s[0:3], s32 offset:116 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v203, off, s[0:3], s32 offset:112 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v204, off, s[0:3], s32 offset:108 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v205, off, s[0:3], s32 offset:104 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v206, off, s[0:3], s32 offset:100 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v207, off, s[0:3], s32 offset:96 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v216, off, s[0:3], s32 offset:92 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v217, off, s[0:3], s32 offset:88 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v218, off, s[0:3], s32 offset:84 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v219, off, s[0:3], s32 offset:80 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v220, off, s[0:3], s32 offset:76 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v221, off, s[0:3], s32 offset:72 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v222, off, s[0:3], s32 offset:68 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v223, off, s[0:3], s32 offset:64 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v232, off, s[0:3], s32 offset:60 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v233, off, s[0:3], s32 offset:56 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v234, off, s[0:3], s32 offset:52 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v235, off, s[0:3], s32 offset:48 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v236, off, s[0:3], s32 offset:44 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v237, off, s[0:3], s32 offset:40 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v238, off, s[0:3], s32 offset:36 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v239, off, s[0:3], s32 offset:32 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v248, off, s[0:3], s32 offset:28 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v249, off, s[0:3], s32 offset:24 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v250, off, s[0:3], s32 offset:20 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v251, off, s[0:3], s32 offset:16 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v252, off, s[0:3], s32 offset:12 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v253, off, s[0:3], s32 offset:8 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v254, off, s[0:3], s32 offset:4 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v255, off, s[0:3], s32 ; 4-byte Folded Spill
+; GCN-NEXT: ;;#ASMSTART
+; GCN-NEXT: ;;#ASMEND
+; GCN-NEXT: buffer_load_dword v255, off, s[0:3], s32 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v254, off, s[0:3], s32 offset:4 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v253, off, s[0:3], s32 offset:8 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v252, off, s[0:3], s32 offset:12 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v251, off, s[0:3], s32 offset:16 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v250, off, s[0:3], s32 offset:20 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v249, off, s[0:3], s32 offset:24 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v248, off, s[0:3], s32 offset:28 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v239, off, s[0:3], s32 offset:32 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v238, off, s[0:3], s32 offset:36 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v237, off, s[0:3], s32 offset:40 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v236, off, s[0:3], s32 offset:44 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v235, off, s[0:3], s32 offset:48 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v234, off, s[0:3], s32 offset:52 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v233, off, s[0:3], s32 offset:56 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v232, off, s[0:3], s32 offset:60 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v223, off, s[0:3], s32 offset:64 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v222, off, s[0:3], s32 offset:68 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v221, off, s[0:3], s32 offset:72 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v220, off, s[0:3], s32 offset:76 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v219, off, s[0:3], s32 offset:80 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v218, off, s[0:3], s32 offset:84 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v217, off, s[0:3], s32 offset:88 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v216, off, s[0:3], s32 offset:92 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v207, off, s[0:3], s32 offset:96 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v206, off, s[0:3], s32 offset:100 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v205, off, s[0:3], s32 offset:104 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v204, off, s[0:3], s32 offset:108 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v203, off, s[0:3], s32 offset:112 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v202, off, s[0:3], s32 offset:116 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v201, off, s[0:3], s32 offset:120 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v200, off, s[0:3], s32 offset:124 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v191, off, s[0:3], s32 offset:128 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v190, off, s[0:3], s32 offset:132 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v189, off, s[0:3], s32 offset:136 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v188, off, s[0:3], s32 offset:140 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v187, off, s[0:3], s32 offset:144 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v186, off, s[0:3], s32 offset:148 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v185, off, s[0:3], s32 offset:152 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v184, off, s[0:3], s32 offset:156 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v175, off, s[0:3], s32 offset:160 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v174, off, s[0:3], s32 offset:164 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v173, off, s[0:3], s32 offset:168 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v172, off, s[0:3], s32 offset:172 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v171, off, s[0:3], s32 offset:176 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v170, off, s[0:3], s32 offset:180 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v169, off, s[0:3], s32 offset:184 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v168, off, s[0:3], s32 offset:188 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v159, off, s[0:3], s32 offset:192 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v158, off, s[0:3], s32 offset:196 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v157, off, s[0:3], s32 offset:200 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v156, off, s[0:3], s32 offset:204 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v155, off, s[0:3], s32 offset:208 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v154, off, s[0:3], s32 offset:212 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v153, off, s[0:3], s32 offset:216 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v152, off, s[0:3], s32 offset:220 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v143, off, s[0:3], s32 offset:224 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v142, off, s[0:3], s32 offset:228 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v141, off, s[0:3], s32 offset:232 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v140, off, s[0:3], s32 offset:236 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v139, off, s[0:3], s32 offset:240 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v138, off, s[0:3], s32 offset:244 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v137, off, s[0:3], s32 offset:248 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v136, off, s[0:3], s32 offset:252 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v127, off, s[0:3], s32 offset:256 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v126, off, s[0:3], s32 offset:260 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v125, off, s[0:3], s32 offset:264 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v124, off, s[0:3], s32 offset:268 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v123, off, s[0:3], s32 offset:272 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v122, off, s[0:3], s32 offset:276 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v121, off, s[0:3], s32 offset:280 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v120, off, s[0:3], s32 offset:284 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v111, off, s[0:3], s32 offset:288 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v110, off, s[0:3], s32 offset:292 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v109, off, s[0:3], s32 offset:296 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v108, off, s[0:3], s32 offset:300 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v107, off, s[0:3], s32 offset:304 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v106, off, s[0:3], s32 offset:308 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v105, off, s[0:3], s32 offset:312 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v104, off, s[0:3], s32 offset:316 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v95, off, s[0:3], s32 offset:320 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v94, off, s[0:3], s32 offset:324 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v93, off, s[0:3], s32 offset:328 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v92, off, s[0:3], s32 offset:332 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v91, off, s[0:3], s32 offset:336 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v90, off, s[0:3], s32 offset:340 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v89, off, s[0:3], s32 offset:344 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v88, off, s[0:3], s32 offset:348 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v79, off, s[0:3], s32 offset:352 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v78, off, s[0:3], s32 offset:356 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v77, off, s[0:3], s32 offset:360 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v76, off, s[0:3], s32 offset:364 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v75, off, s[0:3], s32 offset:368 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v74, off, s[0:3], s32 offset:372 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v73, off, s[0:3], s32 offset:376 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v72, off, s[0:3], s32 offset:380 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v63, off, s[0:3], s32 offset:384 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v62, off, s[0:3], s32 offset:388 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v61, off, s[0:3], s32 offset:392 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v60, off, s[0:3], s32 offset:396 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v59, off, s[0:3], s32 offset:400 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v58, off, s[0:3], s32 offset:404 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v57, off, s[0:3], s32 offset:408 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v56, off, s[0:3], s32 offset:412 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v47, off, s[0:3], s32 offset:416 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v46, off, s[0:3], s32 offset:420 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v45, off, s[0:3], s32 offset:424 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v44, off, s[0:3], s32 offset:428 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v43, off, s[0:3], s32 offset:432 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v42, off, s[0:3], s32 offset:436 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v41, off, s[0:3], s32 offset:440 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v40, off, s[0:3], s32 offset:444 ; 4-byte Folded Reload
+; GCN-NEXT: s_waitcnt vmcnt(0)
+; GCN-NEXT: s_setpc_b64 s[30:31]
call void asm sideeffect "",
"~{v0},~{v1},~{v2},~{v3},~{v4},~{v5},~{v6},~{v7},~{v8},~{v9}
,~{v10},~{v11},~{v12},~{v13},~{v14},~{v15},~{v16},~{v17},~{v18},~{v19}
ret void
}
-; GCN-LABEL: {{^}}spill_sgpr_no_free_vgpr_ipra_tail_call:
-; GCN-NOT: v_writelane_b32
-; GCN-NOT: buffer_store_dword
-; GCN-NOT: swappc
-; GCN-NOT: buffer_load_dword v0, off
-; GCN-NOT: v_readlane_b32
-; GCN: setpc
define void @spill_sgpr_no_free_vgpr_ipra_tail_call() #0 {
+; GCN-LABEL: spill_sgpr_no_free_vgpr_ipra_tail_call:
+; GCN: ; %bb.0:
+; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT: s_getpc_b64 s[4:5]
+; GCN-NEXT: s_add_u32 s4, s4, child_function_ipra_tail_call@rel32@lo+4
+; GCN-NEXT: s_addc_u32 s5, s5, child_function_ipra_tail_call@rel32@hi+12
+; GCN-NEXT: s_setpc_b64 s[4:5]
tail call void @child_function_ipra_tail_call()
ret void
}
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -march=amdgcn -mcpu=verde -enable-misched=0 -post-RA-scheduler=0 -amdgpu-spill-sgpr-to-vgpr=0 < %s | FileCheck -check-prefixes=CHECK,GFX6 %s
; RUN: llc -sgpr-regalloc=basic -vgpr-regalloc=basic -march=amdgcn -mcpu=tonga -enable-misched=0 -post-RA-scheduler=0 -amdgpu-spill-sgpr-to-vgpr=0 < %s | FileCheck --check-prefix=CHECK %s
; RUN: llc -march=amdgcn -mattr=-xnack,+enable-flat-scratch -mcpu=gfx900 -enable-misched=0 -post-RA-scheduler=0 -amdgpu-spill-sgpr-to-vgpr=0 < %s | FileCheck -check-prefixes=CHECK,GFX9-FLATSCR,FLATSCR %s
; mechanism works even when many spills happen.
; Just test that it compiles successfully.
-; CHECK-LABEL: test
-
-; GFX9-FLATSCR: s_mov_b32 [[SOFF1:s[0-9]+]], 4{{$}}
-; GFX9-FLATSCR: scratch_store_dwordx4 off, v[{{[0-9:]+}}], [[SOFF1]] ; 16-byte Folded Spill
-; GFX9-FLATSCR: ;;#ASMSTART
-; GFX9-FLATSCR: s_movk_i32 [[SOFF2:s[0-9]+]], 0x1{{[0-9a-f]+}}{{$}}
-; GFX9-FLATSCR: scratch_load_dwordx4 v[{{[0-9:]+}}], off, [[SOFF2]] ; 16-byte Folded Reload
-
-; GFX10-FLATSCR: scratch_store_dwordx4 off, v[{{[0-9:]+}}], off offset:{{[0-9]+}} ; 16-byte Folded Spill
-; GFX10-FLATSCR: scratch_load_dwordx4 v[{{[0-9:]+}}], off, off offset:{{[0-9]+}} ; 16-byte Folded Reload
define amdgpu_kernel void @test(<1280 x i32> addrspace(1)* %out, <1280 x i32> addrspace(1)* %in) {
+; GFX6-LABEL: test:
+; GFX6: ; %bb.0: ; %entry
+; GFX6-NEXT: s_mov_b32 s44, SCRATCH_RSRC_DWORD0
+; GFX6-NEXT: s_mov_b32 s45, SCRATCH_RSRC_DWORD1
+; GFX6-NEXT: s_mov_b32 s46, -1
+; GFX6-NEXT: s_mov_b32 s47, 0xe8f000
+; GFX6-NEXT: s_add_u32 s44, s44, s3
+; GFX6-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX6-NEXT: v_mbcnt_lo_u32_b32_e64 v0, -1, 0
+; GFX6-NEXT: v_mbcnt_hi_u32_b32_e32 v0, -1, v0
+; GFX6-NEXT: v_lshlrev_b32_e32 v5, 13, v0
+; GFX6-NEXT: s_mov_b32 s18, 0
+; GFX6-NEXT: s_mov_b32 s19, 0xf000
+; GFX6-NEXT: s_waitcnt lgkmcnt(0)
+; GFX6-NEXT: v_add_i32_e32 v0, vcc, s2, v5
+; GFX6-NEXT: v_mov_b32_e32 v1, s3
+; GFX6-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GFX6-NEXT: s_movk_i32 s4, 0x80
+; GFX6-NEXT: s_mov_b32 s5, s18
+; GFX6-NEXT: s_mov_b64 s[6:7], s[18:19]
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[4:7], 0 addr64 offset:3968
+; GFX6-NEXT: s_addc_u32 s45, s45, 0
+; GFX6-NEXT: s_movk_i32 s8, 0x100
+; GFX6-NEXT: s_mov_b32 s9, s18
+; GFX6-NEXT: s_mov_b64 s[10:11], s[18:19]
+; GFX6-NEXT: s_movk_i32 s12, 0x180
+; GFX6-NEXT: s_mov_b32 s13, s18
+; GFX6-NEXT: s_mov_b64 s[14:15], s[18:19]
+; GFX6-NEXT: s_movk_i32 s20, 0x200
+; GFX6-NEXT: s_mov_b32 s21, s18
+; GFX6-NEXT: s_mov_b64 s[22:23], s[18:19]
+; GFX6-NEXT: s_movk_i32 s24, 0x280
+; GFX6-NEXT: s_mov_b32 s25, s18
+; GFX6-NEXT: s_mov_b64 s[26:27], s[18:19]
+; GFX6-NEXT: s_movk_i32 s28, 0x300
+; GFX6-NEXT: s_mov_b32 s29, s18
+; GFX6-NEXT: s_mov_b64 s[30:31], s[18:19]
+; GFX6-NEXT: s_movk_i32 s36, 0x380
+; GFX6-NEXT: s_mov_b32 s37, s18
+; GFX6-NEXT: s_mov_b64 s[38:39], s[18:19]
+; GFX6-NEXT: s_movk_i32 s40, 0x400
+; GFX6-NEXT: s_mov_b32 s41, s18
+; GFX6-NEXT: s_mov_b64 s[42:43], s[18:19]
+; GFX6-NEXT: s_mov_b64 s[16:17], s[2:3]
+; GFX6-NEXT: v_mov_b32_e32 v6, 0
+; GFX6-NEXT: s_mov_b32 s2, 0x3fd00
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:1268 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:1272 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:1276 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:1280 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[4:7], 0 addr64 offset:3984
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:1300 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:1304 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:1308 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:1312 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[4:7], 0 addr64 offset:4000
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:1332 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:1336 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:1340 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:1344 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[4:7], 0 addr64 offset:4016
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:1364 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:1368 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:1372 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:1376 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[4:7], 0 addr64 offset:4032
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:1396 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:1400 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:1404 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:1408 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[4:7], 0 addr64 offset:4048
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:1428 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:1432 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:1436 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:1440 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[4:7], 0 addr64 offset:4064
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:1460 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:1464 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:1468 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:1472 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[4:7], 0 addr64 offset:4080
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:1492 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:1496 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:1500 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:1504 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[8:11], 0 addr64 offset:3968
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:1556 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:1560 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:1564 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:1568 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[8:11], 0 addr64 offset:3984
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:1588 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:1592 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:1596 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:1600 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[8:11], 0 addr64 offset:4000
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:1620 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:1624 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:1628 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:1632 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[8:11], 0 addr64 offset:4016
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:1652 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:1656 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:1660 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:1664 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[8:11], 0 addr64 offset:4032
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:1684 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:1688 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:1692 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:1696 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[8:11], 0 addr64 offset:4048
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:1716 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:1720 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:1724 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:1728 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[8:11], 0 addr64 offset:4064
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:1748 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:1752 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:1756 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:1760 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[8:11], 0 addr64 offset:4080
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:1780 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:1784 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:1788 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:1792 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[12:15], 0 addr64 offset:3968
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:1860 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:1864 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:1868 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:1872 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[12:15], 0 addr64 offset:3984
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:1892 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:1896 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:1900 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:1904 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[12:15], 0 addr64 offset:4000
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:1924 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:1928 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:1932 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:1936 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[12:15], 0 addr64 offset:4016
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:1956 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:1960 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:1964 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:1968 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[12:15], 0 addr64 offset:4032
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:1988 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:1992 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:1996 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:2000 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[12:15], 0 addr64 offset:4048
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:2020 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:2024 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:2028 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:2032 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[12:15], 0 addr64 offset:4064
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:2052 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:2056 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:2060 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:2064 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[12:15], 0 addr64 offset:4080
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:2084 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:2088 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:2092 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:2096 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[20:23], 0 addr64 offset:3968
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:2148 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:2152 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:2156 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:2160 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[20:23], 0 addr64 offset:3984
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:2180 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:2184 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:2188 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:2192 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[20:23], 0 addr64 offset:4000
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:2212 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:2216 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:2220 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:2224 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[20:23], 0 addr64 offset:4016
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:2244 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:2248 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:2252 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:2256 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[20:23], 0 addr64 offset:4032
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:2276 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:2280 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:2284 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:2288 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[20:23], 0 addr64 offset:4048
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:2308 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:2312 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:2316 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:2320 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[20:23], 0 addr64 offset:4064
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:2340 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:2344 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:2348 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:2352 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[20:23], 0 addr64 offset:4080
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:2372 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:2376 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:2380 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:2384 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[24:27], 0 addr64 offset:3968
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:2452 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:2456 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:2460 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:2464 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[24:27], 0 addr64 offset:3984
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:2484 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:2488 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:2492 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:2496 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[24:27], 0 addr64 offset:4000
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:2516 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:2520 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:2524 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:2528 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[24:27], 0 addr64 offset:4016
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:2548 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:2552 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:2556 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:2560 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[24:27], 0 addr64 offset:4032
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:2580 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:2584 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:2588 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:2592 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[24:27], 0 addr64 offset:4048
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:2612 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:2616 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:2620 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:2624 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[24:27], 0 addr64 offset:4064
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:2644 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:2648 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:2652 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:2656 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[24:27], 0 addr64 offset:4080
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:2676 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:2680 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:2684 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:2688 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[28:31], 0 addr64 offset:3968
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:2740 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:2744 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:2748 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:2752 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[28:31], 0 addr64 offset:3984
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:2772 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:2776 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:2780 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:2784 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[28:31], 0 addr64 offset:4000
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:2804 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:2808 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:2812 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:2816 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[28:31], 0 addr64 offset:4016
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:2836 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:2840 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:2844 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:2848 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[28:31], 0 addr64 offset:4032
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:2868 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:2872 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:2876 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:2880 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[28:31], 0 addr64 offset:4048
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:2900 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:2904 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:2908 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:2912 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[28:31], 0 addr64 offset:4064
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:2932 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:2936 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:2940 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:2944 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[28:31], 0 addr64 offset:4080
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:2964 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:2968 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:2972 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:2976 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[36:39], 0 addr64 offset:3968
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:3044 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:3048 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:3052 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:3056 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[36:39], 0 addr64 offset:3984
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:3076 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:3080 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:3084 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:3088 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[36:39], 0 addr64 offset:4000
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:3108 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:3112 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:3116 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:3120 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[36:39], 0 addr64 offset:4016
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:3140 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:3144 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:3148 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:3152 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[36:39], 0 addr64 offset:4032
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:3172 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:3176 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:3180 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:3184 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[36:39], 0 addr64 offset:4048
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:3204 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:3208 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:3212 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:3216 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[36:39], 0 addr64 offset:4064
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:3236 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:3240 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:3244 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:3248 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[36:39], 0 addr64 offset:4080
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:3268 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:3272 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:3276 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:3280 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[40:43], 0 addr64 offset:3968
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:3332 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:3336 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:3340 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:3344 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[40:43], 0 addr64 offset:3984
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:3364 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:3368 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:3372 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:3376 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[40:43], 0 addr64 offset:4000
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:3396 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:3400 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:3404 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:3408 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[40:43], 0 addr64 offset:4016
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:3428 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:3432 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:3436 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:3440 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[40:43], 0 addr64 offset:4032
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:3460 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:3464 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:3468 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:3472 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[40:43], 0 addr64 offset:4048
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:3492 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:3496 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:3500 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:3504 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[7:10], v[0:1], s[40:43], 0 addr64 offset:4064
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v7, off, s[44:47], 0 offset:3524 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v8, off, s[44:47], 0 offset:3528 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v9, off, s[44:47], 0 offset:3532 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v10, off, s[44:47], 0 offset:3536 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[0:1], s[40:43], 0 addr64 offset:4080
+; GFX6-NEXT: s_waitcnt expcnt(3)
+; GFX6-NEXT: v_add_i32_e32 v7, vcc, s0, v5
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:3556 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:3560 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:3564 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:3568 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:16 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:16
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:20 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:24 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:28 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:32 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:32
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:36 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:40 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:44 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:48 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:48
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:52 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:56 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:60 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:64 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:64
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:68 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:72 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:76 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:80 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:80
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:84 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:88 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:92 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:96 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:96
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:100 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:104 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:108 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:112 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:112
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:116 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:120 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:124 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:128 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:128
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:132 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:136 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:140 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:144 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:144
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:148 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:152 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:156 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:160 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:160
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:164 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:168 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:172 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:176 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:176
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:180 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:184 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:188 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:192 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:192
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:196 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:200 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:204 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:208 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:208
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:212 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:216 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:220 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:224 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:224
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:228 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:232 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:236 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:240 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:240
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:244 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:248 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:252 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:256 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:256
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:260 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:264 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:268 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:272 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:272
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:276 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:280 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:284 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:288 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:288
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:292 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:296 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:300 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:304 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:304
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:308 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:312 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:316 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:320 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:320
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:324 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:328 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:332 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:336 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:336
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:340 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:344 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:348 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:352 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:352
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:356 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:360 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:364 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:368 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:368
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:372 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:376 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:380 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:384 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:384
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:388 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:392 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:396 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:400 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:400
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:404 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:408 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:412 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:416 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:416
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:420 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:424 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:428 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:432 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:432
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:436 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:440 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:444 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:448 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:448
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:452 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:456 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:460 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:464 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:464
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:468 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:472 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:476 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:480 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:480
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:484 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:488 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:492 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:496 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:496
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:500 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:504 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:508 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:512 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:512
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:516 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:520 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:524 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:528 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:528
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:532 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:536 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:540 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:544 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:544
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:548 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:552 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:556 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:560 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:560
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:564 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:568 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:572 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:576 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:576
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:580 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:584 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:588 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:592 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:592
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:596 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:600 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:604 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:608 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:608
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:612 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:616 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:620 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:624 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:624
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:628 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:632 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:636 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:640 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:640
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:644 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:648 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:652 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:656 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:656
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:660 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:664 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:668 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:672 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:672
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:676 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:680 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:684 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:688 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:688
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:692 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:696 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:700 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:704 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:704
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:708 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:712 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:716 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:720 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:720
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:724 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:728 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:732 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:736 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:736
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:740 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:744 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:748 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:752 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:752
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:756 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:760 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:764 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:768 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:768
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:772 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:776 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:780 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:784 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:784
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:788 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:792 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:796 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:800 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:800
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:804 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:808 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:812 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:816 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:816
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:820 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:824 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:828 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:832 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:832
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:836 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:840 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:844 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:848 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:848
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:852 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:856 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:860 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:864 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:864
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:868 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:872 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:876 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:880 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:880
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:884 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:888 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:892 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:896 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:896
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:900 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:904 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:908 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:912 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:912
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:916 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:920 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:924 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:928 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:928
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:932 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:936 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:940 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:944 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:944
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:948 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:952 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:956 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:960 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:960
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:964 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:968 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:972 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:976 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:976
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:980 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:984 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:988 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:992 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:992
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:996 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:1000 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:1004 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:1008 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1008
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:1012 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:1016 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:1020 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:1024 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1024
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:1028 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:1032 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:1036 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:1040 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1040
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:1044 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:1048 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:1052 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:1056 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1056
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:1060 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:1064 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:1068 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:1072 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1072
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:1076 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:1080 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:1084 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:1088 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1088
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:1092 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:1096 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:1100 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:1104 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1104
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:1108 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:1112 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:1116 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:1120 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1120
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:1124 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:1128 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:1132 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:1136 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1136
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:1140 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:1144 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:1148 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:1152 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1152
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:1156 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:1160 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:1164 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:1168 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1168
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:1172 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:1176 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:1180 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:1184 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1184
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:1188 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:1192 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:1196 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:1200 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1200
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:1204 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:1208 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:1212 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:1216 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1216
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:1220 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:1224 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:1228 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:1232 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1232
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:1236 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:1240 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:1244 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:1248 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1248
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:1252 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:1256 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:1260 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:1264 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1264
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:1284 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:1288 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:1292 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:1296 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1280
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:1316 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:1320 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:1324 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:1328 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1296
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:1348 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:1352 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:1356 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:1360 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1312
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:1380 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:1384 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:1388 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:1392 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1328
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:1412 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:1416 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:1420 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:1424 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1344
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:1444 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:1448 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:1452 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:1456 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1360
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:1476 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:1480 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:1484 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:1488 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1376
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:1508 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:1512 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:1516 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:1520 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1392
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:1524 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:1528 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:1532 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:1536 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1408
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:1540 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:1544 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:1548 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:1552 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1424
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:1572 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:1576 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:1580 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:1584 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1440
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:1604 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:1608 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:1612 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:1616 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1456
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:1636 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:1640 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:1644 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:1648 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1472
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:1668 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:1672 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:1676 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:1680 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1488
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:1700 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:1704 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:1708 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:1712 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1504
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:1732 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:1736 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:1740 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:1744 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1520
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:1764 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:1768 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:1772 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:1776 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1536
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:1796 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:1800 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:1804 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:1808 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1552
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:1812 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:1816 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:1820 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:1824 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1568
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:1828 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:1832 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:1836 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:1840 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1584
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:1844 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:1848 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:1852 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:1856 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1600
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:1876 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:1880 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:1884 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:1888 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1616
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:1908 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:1912 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:1916 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:1920 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1632
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:1940 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:1944 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:1948 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:1952 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1648
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:1972 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:1976 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:1980 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:1984 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1664
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:2004 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:2008 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:2012 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:2016 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1680
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:2036 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:2040 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:2044 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:2048 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1696
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:2068 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:2072 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:2076 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:2080 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1712
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:2100 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:2104 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:2108 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:2112 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1728
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:2116 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:2120 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:2124 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:2128 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1744
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:2132 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:2136 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:2140 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:2144 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1760
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:2164 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:2168 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:2172 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:2176 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1776
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:2196 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:2200 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:2204 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:2208 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1792
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:2228 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:2232 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:2236 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:2240 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1808
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:2260 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:2264 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:2268 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:2272 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1824
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:2292 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:2296 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:2300 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:2304 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1840
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:2324 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:2328 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:2332 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:2336 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1856
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:2356 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:2360 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:2364 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:2368 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1872
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:2388 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:2392 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:2396 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:2400 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1888
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:2404 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:2408 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:2412 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:2416 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1904
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:2420 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:2424 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:2428 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:2432 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1920
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:2436 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:2440 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:2444 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:2448 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1936
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:2468 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:2472 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:2476 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:2480 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1952
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:2500 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:2504 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:2508 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:2512 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1968
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:2532 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:2536 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:2540 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:2544 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:1984
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:2564 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:2568 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:2572 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:2576 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2000
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:2596 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:2600 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:2604 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:2608 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2016
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:2628 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:2632 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:2636 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:2640 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2032
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:2660 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:2664 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:2668 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:2672 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2048
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:2692 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:2696 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:2700 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:2704 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2064
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:2708 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:2712 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:2716 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:2720 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2080
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:2724 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:2728 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:2732 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:2736 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2096
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:2756 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:2760 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:2764 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:2768 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2112
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:2788 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:2792 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:2796 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:2800 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2128
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:2820 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:2824 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:2828 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:2832 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2144
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:2852 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:2856 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:2860 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:2864 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2160
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:2884 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:2888 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:2892 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:2896 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2176
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:2916 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:2920 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:2924 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:2928 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2192
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:2948 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:2952 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:2956 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:2960 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2208
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:2980 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:2984 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:2988 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:2992 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2224
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:2996 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:3000 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:3004 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:3008 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2240
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:3012 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:3016 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:3020 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:3024 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2256
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:3028 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:3032 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:3036 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:3040 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2272
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:3060 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:3064 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:3068 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:3072 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2288
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:3092 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:3096 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:3100 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:3104 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2304
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:3124 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:3128 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:3132 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:3136 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2320
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:3156 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:3160 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:3164 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:3168 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2336
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:3188 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:3192 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:3196 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:3200 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2352
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:3220 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:3224 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:3228 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:3232 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2368
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:3252 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:3256 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:3260 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:3264 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2384
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:3284 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:3288 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:3292 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:3296 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2400
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:3300 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:3304 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:3308 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:3312 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2416
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:3316 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:3320 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:3324 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:3328 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2432
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:3348 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:3352 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:3356 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:3360 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2448
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:3380 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:3384 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:3388 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:3392 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2464
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:3412 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:3416 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:3420 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:3424 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2480
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:3444 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:3448 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:3452 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:3456 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2496
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:3476 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:3480 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:3484 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:3488 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2512
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:3508 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:3512 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:3516 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:3520 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2528
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:3540 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:3544 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:3548 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:3552 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2544
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:3572 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:3576 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:3580 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:3584 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2560
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:3588 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:3592 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:3596 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:3600 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2576
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:3604 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:3608 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:3612 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:3616 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2592
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:3620 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:3624 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:3628 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:3632 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2608
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:3636 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:3640 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:3644 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:3648 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2624
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:3652 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:3656 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:3660 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:3664 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2640
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:3668 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:3672 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:3676 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:3680 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2656
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:3684 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:3688 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:3692 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:3696 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2672
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:3700 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:3704 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:3708 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:3712 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2688
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:3716 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:3720 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:3724 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:3728 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2704
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:3732 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:3736 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:3740 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:3744 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2720
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:3748 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:3752 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:3756 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:3760 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2736
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:3764 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:3768 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:3772 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:3776 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2752
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:3780 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:3784 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:3788 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:3792 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2768
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:3796 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:3800 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:3804 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:3808 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2784
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:3812 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:3816 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:3820 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:3824 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2800
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:3828 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:3832 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:3836 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:3840 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2816
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:3844 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:3848 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:3852 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:3856 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2832
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:3860 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:3864 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:3868 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:3872 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2848
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:3876 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:3880 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:3884 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:3888 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2864
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:3892 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:3896 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:3900 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:3904 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2880
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:3908 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:3912 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:3916 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:3920 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2896
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:3924 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:3928 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:3932 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:3936 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2912
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:3940 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:3944 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:3948 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:3952 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2928
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:3956 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:3960 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:3964 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:3968 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2944
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:3972 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:3976 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:3980 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:3984 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2960
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:3988 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:3992 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:3996 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:4000 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2976
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:4004 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:4008 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:4012 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:4016 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:2992
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:4020 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:4024 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:4028 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:4032 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3008
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:4036 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:4040 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:4044 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:4048 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3024
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:4052 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:4056 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:4060 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:4064 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3040
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], 0 offset:4068 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], 0 offset:4072 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], 0 offset:4076 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], 0 offset:4080 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3056
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3072
+; GFX6-NEXT: s_mov_b32 s2, 0x40100
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3088
+; GFX6-NEXT: s_mov_b32 s2, 0x40500
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3104
+; GFX6-NEXT: s_mov_b32 s2, 0x40900
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3120
+; GFX6-NEXT: s_mov_b32 s2, 0x40d00
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3136
+; GFX6-NEXT: s_mov_b32 s2, 0x41100
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3152
+; GFX6-NEXT: s_mov_b32 s2, 0x41500
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3168
+; GFX6-NEXT: s_mov_b32 s2, 0x41900
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3184
+; GFX6-NEXT: s_mov_b32 s2, 0x41d00
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3200
+; GFX6-NEXT: s_mov_b32 s2, 0x42100
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3216
+; GFX6-NEXT: s_mov_b32 s2, 0x42500
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3232
+; GFX6-NEXT: s_mov_b32 s2, 0x42900
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3248
+; GFX6-NEXT: s_mov_b32 s2, 0x42d00
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3264
+; GFX6-NEXT: s_mov_b32 s2, 0x43100
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3280
+; GFX6-NEXT: s_mov_b32 s2, 0x43500
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3296
+; GFX6-NEXT: s_mov_b32 s2, 0x43900
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3312
+; GFX6-NEXT: s_mov_b32 s2, 0x43d00
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3328
+; GFX6-NEXT: s_mov_b32 s2, 0x44100
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3344
+; GFX6-NEXT: s_mov_b32 s2, 0x44500
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3360
+; GFX6-NEXT: s_mov_b32 s2, 0x44900
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3376
+; GFX6-NEXT: s_mov_b32 s2, 0x44d00
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3392
+; GFX6-NEXT: s_mov_b32 s2, 0x45100
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3408
+; GFX6-NEXT: s_mov_b32 s2, 0x45500
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3424
+; GFX6-NEXT: s_mov_b32 s2, 0x45900
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3440
+; GFX6-NEXT: s_mov_b32 s2, 0x45d00
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3456
+; GFX6-NEXT: s_mov_b32 s2, 0x46100
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3472
+; GFX6-NEXT: s_mov_b32 s2, 0x46500
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3488
+; GFX6-NEXT: s_mov_b32 s2, 0x46900
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3504
+; GFX6-NEXT: s_mov_b32 s2, 0x46d00
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3520
+; GFX6-NEXT: s_mov_b32 s2, 0x47100
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3536
+; GFX6-NEXT: s_mov_b32 s2, 0x47500
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3552
+; GFX6-NEXT: s_mov_b32 s2, 0x47900
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3568
+; GFX6-NEXT: s_mov_b32 s2, 0x47d00
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3584
+; GFX6-NEXT: s_mov_b32 s2, 0x48100
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3600
+; GFX6-NEXT: s_mov_b32 s2, 0x48500
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3616
+; GFX6-NEXT: s_mov_b32 s2, 0x48900
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3632
+; GFX6-NEXT: s_mov_b32 s2, 0x48d00
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3648
+; GFX6-NEXT: s_mov_b32 s2, 0x49100
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3664
+; GFX6-NEXT: s_mov_b32 s2, 0x49500
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3680
+; GFX6-NEXT: s_mov_b32 s2, 0x49900
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3696
+; GFX6-NEXT: s_mov_b32 s2, 0x49d00
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3712
+; GFX6-NEXT: s_mov_b32 s2, 0x4a100
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3728
+; GFX6-NEXT: s_mov_b32 s2, 0x4a500
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3744
+; GFX6-NEXT: s_mov_b32 s2, 0x4a900
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3760
+; GFX6-NEXT: s_mov_b32 s2, 0x4ad00
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3776
+; GFX6-NEXT: s_mov_b32 s2, 0x4b100
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3792
+; GFX6-NEXT: s_mov_b32 s2, 0x4b500
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3808
+; GFX6-NEXT: s_mov_b32 s2, 0x4b900
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3824
+; GFX6-NEXT: s_mov_b32 s2, 0x4bd00
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3840
+; GFX6-NEXT: s_mov_b32 s2, 0x4c100
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3856
+; GFX6-NEXT: s_mov_b32 s2, 0x4c500
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3872
+; GFX6-NEXT: s_mov_b32 s2, 0x4c900
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3888
+; GFX6-NEXT: s_mov_b32 s2, 0x4cd00
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3904
+; GFX6-NEXT: s_mov_b32 s2, 0x4d100
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3920
+; GFX6-NEXT: s_mov_b32 s2, 0x4d500
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3936
+; GFX6-NEXT: s_mov_b32 s2, 0x4d900
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3952
+; GFX6-NEXT: s_mov_b32 s2, 0x4dd00
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3968
+; GFX6-NEXT: s_mov_b32 s2, 0x4e100
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:3984
+; GFX6-NEXT: s_mov_b32 s2, 0x4e500
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:4000
+; GFX6-NEXT: s_mov_b32 s2, 0x4e900
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:4016
+; GFX6-NEXT: s_mov_b32 s2, 0x4ed00
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:4032
+; GFX6-NEXT: s_mov_b32 s2, 0x4f100
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:4048
+; GFX6-NEXT: s_mov_b32 s2, 0x4f500
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:4064
+; GFX6-NEXT: s_mov_b32 s2, 0x4f900
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[44:47], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[44:47], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[44:47], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[44:47], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[5:6], s[16:19], 0 addr64 offset:4080
+; GFX6-NEXT: ;;#ASMSTART
+; GFX6-NEXT: ;;#ASMEND
+; GFX6-NEXT: ;;#ASMSTART
+; GFX6-NEXT: ;;#ASMEND
+; GFX6-NEXT: ;;#ASMSTART
+; GFX6-NEXT: ;;#ASMEND
+; GFX6-NEXT: ;;#ASMSTART
+; GFX6-NEXT: ;;#ASMEND
+; GFX6-NEXT: ;;#ASMSTART
+; GFX6-NEXT: ;;#ASMEND
+; GFX6-NEXT: ;;#ASMSTART
+; GFX6-NEXT: ;;#ASMEND
+; GFX6-NEXT: ;;#ASMSTART
+; GFX6-NEXT: ;;#ASMEND
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:3556 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:3560 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:3564 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:3568 ; 4-byte Folded Reload
+; GFX6-NEXT: v_mov_b32_e32 v4, s1
+; GFX6-NEXT: v_addc_u32_e32 v8, vcc, 0, v4, vcc
+; GFX6-NEXT: s_mov_b64 s[2:3], s[18:19]
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[40:43], 0 addr64 offset:4080
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:3524 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:3528 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:3532 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:3536 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[40:43], 0 addr64 offset:4064
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:3492 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:3496 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:3500 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:3504 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[40:43], 0 addr64 offset:4048
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:3460 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:3464 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:3468 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:3472 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[40:43], 0 addr64 offset:4032
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:3428 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:3432 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:3436 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:3440 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[40:43], 0 addr64 offset:4016
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:3396 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:3400 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:3404 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:3408 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[40:43], 0 addr64 offset:4000
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:3364 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:3368 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:3372 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:3376 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[40:43], 0 addr64 offset:3984
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:3332 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:3336 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:3340 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:3344 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[40:43], 0 addr64 offset:3968
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:3268 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:3272 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:3276 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:3280 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[36:39], 0 addr64 offset:4080
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:3236 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:3240 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:3244 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:3248 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[36:39], 0 addr64 offset:4064
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:3204 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:3208 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:3212 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:3216 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[36:39], 0 addr64 offset:4048
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:3172 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:3176 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:3180 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:3184 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[36:39], 0 addr64 offset:4032
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:3140 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:3144 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:3148 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:3152 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[36:39], 0 addr64 offset:4016
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:3108 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:3112 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:3116 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:3120 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[36:39], 0 addr64 offset:4000
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:3076 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:3080 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:3084 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:3088 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[36:39], 0 addr64 offset:3984
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:3044 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:3048 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:3052 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:3056 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[36:39], 0 addr64 offset:3968
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:2964 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:2968 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:2972 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:2976 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[28:31], 0 addr64 offset:4080
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:2932 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:2936 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:2940 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:2944 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[28:31], 0 addr64 offset:4064
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:2900 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:2904 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:2908 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:2912 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[28:31], 0 addr64 offset:4048
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:2868 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:2872 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:2876 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:2880 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[28:31], 0 addr64 offset:4032
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:2836 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:2840 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:2844 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:2848 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[28:31], 0 addr64 offset:4016
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:2804 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:2808 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:2812 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:2816 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[28:31], 0 addr64 offset:4000
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:2772 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:2776 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:2780 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:2784 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[28:31], 0 addr64 offset:3984
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:2740 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:2744 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:2748 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:2752 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[28:31], 0 addr64 offset:3968
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:2676 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:2680 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:2684 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:2688 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[24:27], 0 addr64 offset:4080
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:2644 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:2648 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:2652 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:2656 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[24:27], 0 addr64 offset:4064
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:2612 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:2616 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:2620 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:2624 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[24:27], 0 addr64 offset:4048
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:2580 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:2584 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:2588 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:2592 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[24:27], 0 addr64 offset:4032
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:2548 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:2552 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:2556 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:2560 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[24:27], 0 addr64 offset:4016
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:2516 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:2520 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:2524 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:2528 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[24:27], 0 addr64 offset:4000
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:2484 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:2488 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:2492 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:2496 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[24:27], 0 addr64 offset:3984
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:2452 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:2456 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:2460 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:2464 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[24:27], 0 addr64 offset:3968
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:2372 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:2376 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:2380 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:2384 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[20:23], 0 addr64 offset:4080
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:2340 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:2344 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:2348 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:2352 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[20:23], 0 addr64 offset:4064
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:2308 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:2312 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:2316 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:2320 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[20:23], 0 addr64 offset:4048
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:2276 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:2280 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:2284 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:2288 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[20:23], 0 addr64 offset:4032
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:2244 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:2248 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:2252 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:2256 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[20:23], 0 addr64 offset:4016
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:2212 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:2216 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:2220 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:2224 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[20:23], 0 addr64 offset:4000
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:2180 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:2184 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:2188 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:2192 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[20:23], 0 addr64 offset:3984
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:2148 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:2152 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:2156 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:2160 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[20:23], 0 addr64 offset:3968
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:2084 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:2088 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:2092 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:2096 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[12:15], 0 addr64 offset:4080
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:2052 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:2056 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:2060 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:2064 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[12:15], 0 addr64 offset:4064
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:2020 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:2024 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:2028 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:2032 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[12:15], 0 addr64 offset:4048
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:1988 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:1992 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:1996 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:2000 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[12:15], 0 addr64 offset:4032
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:1956 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:1960 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:1964 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:1968 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[12:15], 0 addr64 offset:4016
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:1924 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:1928 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:1932 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:1936 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[12:15], 0 addr64 offset:4000
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:1892 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:1896 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:1900 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:1904 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[12:15], 0 addr64 offset:3984
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:1860 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:1864 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:1868 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:1872 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[12:15], 0 addr64 offset:3968
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:1780 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:1784 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:1788 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:1792 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[8:11], 0 addr64 offset:4080
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:1748 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:1752 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:1756 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:1760 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[8:11], 0 addr64 offset:4064
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:1716 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:1720 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:1724 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:1728 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[8:11], 0 addr64 offset:4048
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:1684 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:1688 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:1692 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:1696 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[8:11], 0 addr64 offset:4032
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:1652 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:1656 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:1660 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:1664 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[8:11], 0 addr64 offset:4016
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:1620 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:1624 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:1628 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:1632 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[8:11], 0 addr64 offset:4000
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:1588 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:1592 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:1596 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:1600 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[8:11], 0 addr64 offset:3984
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:1556 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:1560 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:1564 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:1568 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[8:11], 0 addr64 offset:3968
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:1492 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:1496 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:1500 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:1504 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[4:7], 0 addr64 offset:4080
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:1460 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:1464 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:1468 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:1472 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[4:7], 0 addr64 offset:4064
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:1428 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:1432 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:1436 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:1440 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[4:7], 0 addr64 offset:4048
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:1396 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:1400 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:1404 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:1408 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[4:7], 0 addr64 offset:4032
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:1364 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:1368 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:1372 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:1376 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[4:7], 0 addr64 offset:4016
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:1332 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:1336 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:1340 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:1344 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[4:7], 0 addr64 offset:4000
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:1300 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:1304 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:1308 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:1312 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[4:7], 0 addr64 offset:3984
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[44:47], 0 offset:1268 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v10, off, s[44:47], 0 offset:1272 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v11, off, s[44:47], 0 offset:1276 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v12, off, s[44:47], 0 offset:1280 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[9:12], v[7:8], s[4:7], 0 addr64 offset:3968
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:4080
+; GFX6-NEXT: s_mov_b32 s4, 0x4f900
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x4f500
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:4064
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x4f100
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:4048
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x4ed00
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:4032
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x4e900
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:4016
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x4e500
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:4000
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x4e100
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3984
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x4dd00
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3968
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x4d900
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3952
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x4d500
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3936
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x4d100
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3920
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x4cd00
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3904
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x4c900
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3888
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x4c500
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3872
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x4c100
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3856
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x4bd00
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3840
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x4b900
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3824
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x4b500
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3808
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x4b100
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3792
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x4ad00
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3776
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x4a900
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3760
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x4a500
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3744
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x4a100
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3728
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x49d00
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3712
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x49900
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3696
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x49500
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3680
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x49100
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3664
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x48d00
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3648
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x48900
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3632
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x48500
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3616
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x48100
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3600
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x47d00
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3584
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x47900
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3568
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x47500
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3552
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x47100
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3536
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x46d00
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3520
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x46900
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3504
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x46500
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3488
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x46100
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3472
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x45d00
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3456
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x45900
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3440
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x45500
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3424
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x45100
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3408
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x44d00
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3392
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x44900
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3376
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x44500
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3360
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x44100
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3344
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x43d00
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3328
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x43900
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3312
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x43500
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3296
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x43100
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3280
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x42d00
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3264
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x42900
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3248
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x42500
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3232
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x42100
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3216
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x41d00
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3200
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x41900
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3184
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x41500
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3168
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x41100
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3152
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x40d00
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3136
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x40900
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3120
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x40500
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3104
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x40100
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3088
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x3fd00
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3072
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3056
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:4068 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:4072 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:4076 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:4080 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3040
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:4052 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:4056 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:4060 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:4064 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3024
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:4036 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:4040 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:4044 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:4048 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:3008
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:4020 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:4024 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:4028 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:4032 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2992
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:4004 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:4008 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:4012 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:4016 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2976
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:3988 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:3992 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:3996 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:4000 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2960
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:3972 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:3976 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:3980 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:3984 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2944
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:3956 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:3960 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:3964 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:3968 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2928
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:3940 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:3944 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:3948 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:3952 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2912
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:3924 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:3928 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:3932 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:3936 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2896
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:3908 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:3912 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:3916 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:3920 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2880
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:3892 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:3896 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:3900 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:3904 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2864
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:3876 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:3880 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:3884 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:3888 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2848
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:3860 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:3864 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:3868 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:3872 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2832
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:3844 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:3848 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:3852 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:3856 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2816
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:3828 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:3832 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:3836 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:3840 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2800
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:3812 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:3816 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:3820 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:3824 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2784
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:3796 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:3800 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:3804 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:3808 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2768
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:3780 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:3784 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:3788 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:3792 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2752
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:3764 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:3768 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:3772 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:3776 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2736
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:3748 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:3752 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:3756 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:3760 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2720
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:3732 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:3736 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:3740 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:3744 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2704
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:3716 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:3720 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:3724 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:3728 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2688
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:3700 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:3704 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:3708 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:3712 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2672
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:3684 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:3688 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:3692 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:3696 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2656
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:3668 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:3672 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:3676 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:3680 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2640
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:3652 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:3656 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:3660 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:3664 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2624
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:3636 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:3640 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:3644 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:3648 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2608
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:3620 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:3624 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:3628 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:3632 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2592
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:3604 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:3608 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:3612 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:3616 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2576
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:3588 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:3592 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:3596 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:3600 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2560
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:3572 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:3576 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:3580 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:3584 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2544
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:3540 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:3544 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:3548 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:3552 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2528
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:3508 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:3512 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:3516 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:3520 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2512
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:3476 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:3480 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:3484 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:3488 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2496
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:3444 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:3448 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:3452 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:3456 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2480
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:3412 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:3416 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:3420 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:3424 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2464
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:3380 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:3384 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:3388 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:3392 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2448
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:3348 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:3352 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:3356 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:3360 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2432
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:3316 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:3320 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:3324 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:3328 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2416
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:3300 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:3304 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:3308 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:3312 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2400
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:3284 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:3288 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:3292 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:3296 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2384
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:3252 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:3256 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:3260 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:3264 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2368
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:3220 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:3224 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:3228 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:3232 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2352
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:3188 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:3192 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:3196 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:3200 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2336
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:3156 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:3160 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:3164 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:3168 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2320
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:3124 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:3128 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:3132 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:3136 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2304
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:3092 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:3096 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:3100 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:3104 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2288
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:3060 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:3064 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:3068 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:3072 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2272
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:3028 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:3032 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:3036 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:3040 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2256
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:3012 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:3016 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:3020 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:3024 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2240
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:2996 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:3000 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:3004 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:3008 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2224
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:2980 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:2984 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:2988 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:2992 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2208
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:2948 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:2952 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:2956 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:2960 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2192
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:2916 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:2920 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:2924 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:2928 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2176
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:2884 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:2888 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:2892 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:2896 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2160
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:2852 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:2856 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:2860 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:2864 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2144
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:2820 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:2824 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:2828 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:2832 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2128
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:2788 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:2792 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:2796 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:2800 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2112
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:2756 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:2760 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:2764 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:2768 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2096
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:2724 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:2728 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:2732 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:2736 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2080
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:2708 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:2712 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:2716 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:2720 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2064
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:2692 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:2696 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:2700 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:2704 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2048
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:2660 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:2664 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:2668 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:2672 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2032
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:2628 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:2632 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:2636 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:2640 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2016
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:2596 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:2600 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:2604 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:2608 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:2000
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:2564 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:2568 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:2572 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:2576 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1984
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:2532 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:2536 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:2540 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:2544 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1968
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:2500 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:2504 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:2508 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:2512 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1952
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:2468 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:2472 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:2476 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:2480 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1936
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:2436 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:2440 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:2444 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:2448 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1920
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:2420 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:2424 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:2428 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:2432 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1904
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:2404 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:2408 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:2412 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:2416 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1888
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:2388 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:2392 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:2396 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:2400 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1872
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:2356 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:2360 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:2364 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:2368 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1856
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:2324 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:2328 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:2332 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:2336 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1840
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:2292 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:2296 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:2300 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:2304 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1824
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:2260 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:2264 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:2268 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:2272 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1808
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:2228 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:2232 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:2236 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:2240 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1792
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:2196 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:2200 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:2204 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:2208 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1776
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:2164 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:2168 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:2172 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:2176 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1760
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:2132 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:2136 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:2140 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:2144 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1744
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:2116 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:2120 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:2124 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:2128 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1728
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:2100 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:2104 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:2108 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:2112 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1712
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:2068 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:2072 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:2076 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:2080 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1696
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:2036 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:2040 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:2044 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:2048 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1680
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:2004 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:2008 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:2012 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:2016 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1664
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:1972 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:1976 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:1980 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:1984 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1648
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:1940 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:1944 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:1948 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:1952 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1632
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:1908 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:1912 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:1916 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:1920 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1616
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:1876 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:1880 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:1884 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:1888 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1600
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:1844 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:1848 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:1852 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:1856 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1584
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:1828 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:1832 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:1836 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:1840 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1568
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:1812 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:1816 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:1820 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:1824 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1552
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:1796 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:1800 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:1804 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:1808 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1536
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:1764 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:1768 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:1772 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:1776 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1520
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:1732 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:1736 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:1740 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:1744 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1504
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:1700 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:1704 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:1708 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:1712 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1488
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:1668 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:1672 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:1676 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:1680 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1472
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:1636 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:1640 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:1644 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:1648 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1456
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:1604 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:1608 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:1612 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:1616 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1440
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:1572 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:1576 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:1580 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:1584 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1424
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:1540 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:1544 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:1548 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:1552 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1408
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:1524 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:1528 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:1532 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:1536 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1392
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:1508 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:1512 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:1516 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:1520 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1376
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:1476 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:1480 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:1484 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:1488 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1360
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:1444 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:1448 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:1452 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:1456 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1344
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:1412 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:1416 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:1420 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:1424 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1328
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:1380 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:1384 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:1388 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:1392 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1312
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:1348 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:1352 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:1356 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:1360 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1296
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:1316 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:1320 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:1324 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:1328 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1280
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:1284 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:1288 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:1292 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:1296 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1264
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:1252 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:1256 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:1260 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:1264 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1248
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:1236 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:1240 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:1244 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:1248 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1232
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:1220 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:1224 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:1228 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:1232 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1216
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:1204 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:1208 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:1212 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:1216 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1200
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:1188 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:1192 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:1196 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:1200 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1184
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:1172 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:1176 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:1180 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:1184 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1168
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:1156 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:1160 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:1164 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:1168 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1152
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:1140 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:1144 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:1148 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:1152 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1136
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:1124 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:1128 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:1132 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:1136 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1120
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:1108 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:1112 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:1116 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:1120 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1104
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:1092 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:1096 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:1100 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:1104 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1088
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:1076 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:1080 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:1084 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:1088 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1072
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:1060 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:1064 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:1068 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:1072 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1056
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:1044 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:1048 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:1052 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:1056 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1040
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:1028 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:1032 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:1036 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:1040 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1024
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:1012 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:1016 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:1020 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:1024 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:1008
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:996 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:1000 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:1004 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:1008 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:992
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:980 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:984 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:988 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:992 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:976
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:964 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:968 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:972 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:976 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:960
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:948 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:952 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:956 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:960 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:944
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:932 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:936 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:940 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:944 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:928
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:916 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:920 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:924 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:928 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:912
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:900 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:904 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:908 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:912 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:896
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:884 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:888 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:892 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:896 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:880
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:868 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:872 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:876 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:880 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:864
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:852 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:856 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:860 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:864 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:848
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:836 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:840 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:844 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:848 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:832
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:820 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:824 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:828 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:832 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:816
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:804 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:808 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:812 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:816 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:800
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:788 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:792 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:796 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:800 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:784
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:772 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:776 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:780 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:784 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:768
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:756 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:760 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:764 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:768 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:752
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:740 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:744 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:748 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:752 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:736
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:724 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:728 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:732 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:736 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:720
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:708 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:712 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:716 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:720 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:704
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:692 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:696 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:700 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:704 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:688
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:676 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:680 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:684 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:688 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:672
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:660 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:664 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:668 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:672 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:656
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:644 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:648 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:652 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:656 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:640
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:628 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:632 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:636 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:640 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:624
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:612 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:616 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:620 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:624 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:608
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:596 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:600 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:604 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:608 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:592
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:580 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:584 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:588 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:592 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:576
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:564 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:568 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:572 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:576 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:560
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:548 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:552 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:556 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:560 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:544
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:532 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:536 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:540 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:544 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:528
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:516 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:520 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:524 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:528 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:512
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:500 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:504 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:508 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:512 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:496
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:484 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:488 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:492 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:496 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:480
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:468 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:472 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:476 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:480 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:464
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:452 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:456 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:460 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:464 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:448
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:436 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:440 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:444 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:448 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:432
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:420 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:424 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:428 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:432 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:416
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:404 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:408 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:412 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:416 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:400
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:388 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:392 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:396 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:400 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:384
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:372 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:376 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:380 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:384 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:368
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:356 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:360 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:364 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:368 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:352
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:340 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:344 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:348 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:352 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:336
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:324 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:328 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:332 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:336 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:320
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:308 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:312 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:316 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:320 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:304
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:292 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:296 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:300 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:304 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:288
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:276 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:280 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:284 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:288 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:272
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:260 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:264 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:268 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:272 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:256
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:244 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:248 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:252 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:256 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:240
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:228 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:232 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:236 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:240 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:224
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:212 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:216 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:220 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:224 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:208
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:196 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:200 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:204 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:208 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:192
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:180 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:184 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:188 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:192 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:176
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:164 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:168 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:172 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:176 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:160
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:148 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:152 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:156 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:160 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:144
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:132 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:136 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:140 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:144 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:128
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:116 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:120 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:124 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:128 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:112
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:100 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:104 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:108 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:112 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:96
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:84 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:88 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:92 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:96 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:80
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:68 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:72 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:76 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:80 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:64
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:52 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:56 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:60 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:64 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:48
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:36 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:40 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:44 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:48 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:32
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:20 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:24 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:28 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:32 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64 offset:16
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v0, off, s[44:47], 0 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[44:47], 0 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[44:47], 0 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[44:47], 0 offset:16 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[5:6], s[0:3], 0 addr64
+; GFX6-NEXT: s_endpgm
+;
+; GFX9-FLATSCR-LABEL: test:
+; GFX9-FLATSCR: ; %bb.0: ; %entry
+; GFX9-FLATSCR-NEXT: s_add_u32 flat_scratch_lo, s2, s5
+; GFX9-FLATSCR-NEXT: s_addc_u32 flat_scratch_hi, s3, 0
+; GFX9-FLATSCR-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX9-FLATSCR-NEXT: v_mbcnt_lo_u32_b32 v0, -1, 0
+; GFX9-FLATSCR-NEXT: v_mbcnt_hi_u32_b32 v0, -1, v0
+; GFX9-FLATSCR-NEXT: v_lshlrev_b32_e32 v5, 13, v0
+; GFX9-FLATSCR-NEXT: s_movk_i32 s4, 0x80
+; GFX9-FLATSCR-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-FLATSCR-NEXT: v_add_co_u32_e32 v2, vcc, s2, v5
+; GFX9-FLATSCR-NEXT: v_mov_b32_e32 v0, s3
+; GFX9-FLATSCR-NEXT: v_addc_co_u32_e32 v3, vcc, 0, v0, vcc
+; GFX9-FLATSCR-NEXT: v_add_co_u32_e32 v0, vcc, s4, v2
+; GFX9-FLATSCR-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v3, vcc
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:3968
+; GFX9-FLATSCR-NEXT: s_mov_b32 s4, 4
+; GFX9-FLATSCR-NEXT: s_movk_i32 s5, 0x84
+; GFX9-FLATSCR-NEXT: s_movk_i32 s6, 0x104
+; GFX9-FLATSCR-NEXT: s_movk_i32 s7, 0x184
+; GFX9-FLATSCR-NEXT: s_movk_i32 s8, 0x204
+; GFX9-FLATSCR-NEXT: s_movk_i32 s9, 0x284
+; GFX9-FLATSCR-NEXT: s_movk_i32 s10, 0x304
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x384
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:3984
+; GFX9-FLATSCR-NEXT: s_mov_b32 s4, 20
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:4000
+; GFX9-FLATSCR-NEXT: s_mov_b32 s4, 36
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:4016
+; GFX9-FLATSCR-NEXT: s_mov_b32 s4, 52
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:4032
+; GFX9-FLATSCR-NEXT: s_movk_i32 s4, 0x44
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:4048
+; GFX9-FLATSCR-NEXT: s_movk_i32 s4, 0x54
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:4064
+; GFX9-FLATSCR-NEXT: s_movk_i32 s4, 0x64
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:4080
+; GFX9-FLATSCR-NEXT: s_movk_i32 s4, 0x74
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: s_movk_i32 s4, 0x100
+; GFX9-FLATSCR-NEXT: v_add_co_u32_e32 v0, vcc, s4, v2
+; GFX9-FLATSCR-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v3, vcc
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:3968
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s5 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:3984
+; GFX9-FLATSCR-NEXT: s_movk_i32 s5, 0x94
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s5 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:4000
+; GFX9-FLATSCR-NEXT: s_movk_i32 s5, 0xa4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s5 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:4016
+; GFX9-FLATSCR-NEXT: s_movk_i32 s5, 0xb4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s5 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:4032
+; GFX9-FLATSCR-NEXT: s_movk_i32 s5, 0xc4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s5 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:4048
+; GFX9-FLATSCR-NEXT: s_movk_i32 s5, 0xd4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s5 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:4064
+; GFX9-FLATSCR-NEXT: s_movk_i32 s5, 0xe4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s5 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:4080
+; GFX9-FLATSCR-NEXT: s_movk_i32 s5, 0xf4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s5 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: s_movk_i32 s5, 0x180
+; GFX9-FLATSCR-NEXT: v_add_co_u32_e32 v0, vcc, s5, v2
+; GFX9-FLATSCR-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v3, vcc
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:3968
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s6 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:3984
+; GFX9-FLATSCR-NEXT: s_movk_i32 s6, 0x114
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s6 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:4000
+; GFX9-FLATSCR-NEXT: s_movk_i32 s6, 0x124
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s6 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:4016
+; GFX9-FLATSCR-NEXT: s_movk_i32 s6, 0x134
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s6 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:4032
+; GFX9-FLATSCR-NEXT: s_movk_i32 s6, 0x144
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s6 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:4048
+; GFX9-FLATSCR-NEXT: s_movk_i32 s6, 0x154
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s6 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:4064
+; GFX9-FLATSCR-NEXT: s_movk_i32 s6, 0x164
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s6 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:4080
+; GFX9-FLATSCR-NEXT: s_movk_i32 s6, 0x174
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s6 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: s_movk_i32 s6, 0x200
+; GFX9-FLATSCR-NEXT: v_add_co_u32_e32 v0, vcc, s6, v2
+; GFX9-FLATSCR-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v3, vcc
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:3968
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s7 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:3984
+; GFX9-FLATSCR-NEXT: s_movk_i32 s7, 0x194
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s7 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:4000
+; GFX9-FLATSCR-NEXT: s_movk_i32 s7, 0x1a4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s7 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:4016
+; GFX9-FLATSCR-NEXT: s_movk_i32 s7, 0x1b4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s7 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:4032
+; GFX9-FLATSCR-NEXT: s_movk_i32 s7, 0x1c4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s7 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:4048
+; GFX9-FLATSCR-NEXT: s_movk_i32 s7, 0x1d4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s7 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:4064
+; GFX9-FLATSCR-NEXT: s_movk_i32 s7, 0x1e4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s7 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:4080
+; GFX9-FLATSCR-NEXT: s_movk_i32 s7, 0x1f4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s7 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: s_movk_i32 s7, 0x280
+; GFX9-FLATSCR-NEXT: v_add_co_u32_e32 v0, vcc, s7, v2
+; GFX9-FLATSCR-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v3, vcc
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:3968
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s8 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:3984
+; GFX9-FLATSCR-NEXT: s_movk_i32 s8, 0x214
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s8 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:4000
+; GFX9-FLATSCR-NEXT: s_movk_i32 s8, 0x224
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s8 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:4016
+; GFX9-FLATSCR-NEXT: s_movk_i32 s8, 0x234
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s8 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:4032
+; GFX9-FLATSCR-NEXT: s_movk_i32 s8, 0x244
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s8 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:4048
+; GFX9-FLATSCR-NEXT: s_movk_i32 s8, 0x254
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s8 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:4064
+; GFX9-FLATSCR-NEXT: s_movk_i32 s8, 0x264
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s8 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:4080
+; GFX9-FLATSCR-NEXT: s_movk_i32 s8, 0x274
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s8 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: s_movk_i32 s8, 0x300
+; GFX9-FLATSCR-NEXT: v_add_co_u32_e32 v0, vcc, s8, v2
+; GFX9-FLATSCR-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v3, vcc
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:3968
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s9 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:3984
+; GFX9-FLATSCR-NEXT: s_movk_i32 s9, 0x294
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s9 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:4000
+; GFX9-FLATSCR-NEXT: s_movk_i32 s9, 0x2a4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s9 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:4016
+; GFX9-FLATSCR-NEXT: s_movk_i32 s9, 0x2b4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s9 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:4032
+; GFX9-FLATSCR-NEXT: s_movk_i32 s9, 0x2c4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s9 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:4048
+; GFX9-FLATSCR-NEXT: s_movk_i32 s9, 0x2d4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s9 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:4064
+; GFX9-FLATSCR-NEXT: s_movk_i32 s9, 0x2e4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s9 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:4080
+; GFX9-FLATSCR-NEXT: s_movk_i32 s9, 0x2f4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s9 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: s_movk_i32 s9, 0x380
+; GFX9-FLATSCR-NEXT: v_add_co_u32_e32 v0, vcc, s9, v2
+; GFX9-FLATSCR-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v3, vcc
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:3968
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s10 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:3984
+; GFX9-FLATSCR-NEXT: s_movk_i32 s10, 0x314
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s10 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:4000
+; GFX9-FLATSCR-NEXT: s_movk_i32 s10, 0x324
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s10 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:4016
+; GFX9-FLATSCR-NEXT: s_movk_i32 s10, 0x334
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s10 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:4032
+; GFX9-FLATSCR-NEXT: s_movk_i32 s10, 0x344
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s10 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:4048
+; GFX9-FLATSCR-NEXT: s_movk_i32 s10, 0x354
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s10 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:4064
+; GFX9-FLATSCR-NEXT: s_movk_i32 s10, 0x364
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s10 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:4080
+; GFX9-FLATSCR-NEXT: s_movk_i32 s10, 0x374
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s10 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: s_movk_i32 s10, 0x400
+; GFX9-FLATSCR-NEXT: v_add_co_u32_e32 v0, vcc, s10, v2
+; GFX9-FLATSCR-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v3, vcc
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:3968
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:3984
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x394
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:4000
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x3a4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:4016
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x3b4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:4032
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x3c4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:4048
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x3d4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:4064
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v[0:1], off offset:4080
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x3e4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(1)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x3f4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(1)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3]
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x404
+; GFX9-FLATSCR-NEXT: v_mov_b32_e32 v6, s1
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:16
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x414
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:32
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x424
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:48
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x434
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:64
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x444
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:80
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x454
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:96
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x464
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:112
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x474
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:128
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x484
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:144
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x494
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:160
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x4a4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:176
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x4b4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:192
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x4c4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:208
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x4d4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:224
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x4e4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:240
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x4f4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:256
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x504
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:272
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x514
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:288
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x524
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:304
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x534
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:320
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x544
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:336
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x554
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:352
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x564
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:368
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x574
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:384
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x584
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:400
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x594
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:416
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x5a4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:432
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x5b4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:448
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x5c4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:464
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x5d4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:480
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x5e4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:496
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x5f4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:512
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x604
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:528
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x614
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:544
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x624
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:560
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x634
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:576
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x644
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:592
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x654
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:608
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x664
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:624
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x674
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:640
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x684
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:656
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x694
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:672
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x6a4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:688
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x6b4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:704
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x6c4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:720
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x6d4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:736
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x6e4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:752
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x6f4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:768
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x704
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:784
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x714
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:800
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x724
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:816
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x734
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:832
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x744
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:848
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x754
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:864
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x764
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:880
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x774
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:896
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x784
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:912
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x794
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:928
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x7a4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:944
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x7b4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:960
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x7c4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:976
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x7d4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:992
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x7e4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1008
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x7f4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1024
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x804
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1040
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x814
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1056
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x824
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1072
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x834
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1088
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x844
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1104
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x854
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1120
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x864
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1136
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x874
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1152
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x884
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1168
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x894
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1184
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x8a4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1200
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x8b4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1216
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x8c4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1232
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x8d4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1248
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x8e4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1264
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x8f4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1280
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x904
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1296
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x914
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1312
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x924
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1328
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x934
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1344
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x944
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1360
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x954
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1376
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x964
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1392
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x974
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1408
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x984
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1424
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x994
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1440
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x9a4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1456
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x9b4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1472
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x9c4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1488
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x9d4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1504
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x9e4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1520
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x9f4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1536
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xa04
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1552
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xa14
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1568
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xa24
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1584
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xa34
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1600
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xa44
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1616
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xa54
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1632
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xa64
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1648
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xa74
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1664
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xa84
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1680
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xa94
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1696
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xaa4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1712
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xab4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1728
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xac4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1744
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xad4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1760
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xae4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1776
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xaf4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1792
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xb04
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1808
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xb14
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1824
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xb24
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1840
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xb34
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1856
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xb44
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1872
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xb54
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1888
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xb64
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1904
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xb74
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1920
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xb84
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1936
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xb94
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1952
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xba4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1968
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xbb4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1984
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xbc4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2000
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xbd4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2016
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xbe4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2032
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xbf4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2048
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xc04
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2064
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xc14
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2080
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xc24
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2096
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xc34
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2112
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xc44
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2128
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xc54
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2144
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xc64
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2160
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xc74
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2176
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xc84
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2192
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xc94
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2208
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xca4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2224
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xcb4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2240
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xcc4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2256
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xcd4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2272
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xce4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2288
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xcf4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2304
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xd04
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2320
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xd14
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2336
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xd24
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2352
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xd34
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2368
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xd44
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2384
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xd54
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2400
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xd64
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2416
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xd74
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2432
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xd84
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2448
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xd94
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2464
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xda4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2480
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xdb4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2496
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xdc4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2512
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xdd4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2528
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xde4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2544
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xdf4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2560
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xe04
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2576
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xe14
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2592
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xe24
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2608
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xe34
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2624
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xe44
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2640
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xe54
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2656
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xe64
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2672
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xe74
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2688
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xe84
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2704
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xe94
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2720
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xea4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2736
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xeb4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2752
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xec4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2768
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xed4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2784
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xee4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2800
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xef4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2816
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xf04
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2832
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xf14
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2848
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xf24
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2864
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xf34
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2880
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xf44
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2896
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xf54
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2912
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xf64
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2928
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xf74
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2944
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xf84
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2960
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xf94
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2976
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xfa4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2992
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xfb4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3008
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xfc4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3024
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xfd4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3040
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xfe4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3056
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0xff4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3072
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x1004
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3088
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x1014
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3104
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x1024
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3120
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x1034
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3136
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x1044
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3152
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x1054
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3168
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x1064
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3184
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x1074
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3200
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x1084
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3216
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x1094
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3232
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x10a4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3248
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x10b4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3264
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x10c4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3280
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x10d4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3296
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x10e4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3312
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x10f4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3328
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x1104
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3344
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x1114
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3360
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x1124
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3376
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x1134
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3392
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x1144
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3408
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x1154
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3424
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x1164
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3440
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x1174
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3456
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x1184
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3472
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x1194
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3488
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x11a4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3504
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x11b4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3520
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x11c4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3536
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x11d4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3552
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x11e4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3568
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x11f4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3584
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x1204
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3600
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x1214
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3616
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x1224
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3632
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x1234
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3648
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x1244
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3664
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x1254
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3680
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x1264
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3696
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x1274
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3712
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x1284
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3728
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x1294
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3744
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x12a4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3760
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x12b4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3776
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x12c4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3792
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x12d4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3808
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x12e4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3824
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x12f4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3840
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x1304
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3856
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x1314
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3872
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x1324
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3888
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x1334
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3904
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x1344
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3920
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x1354
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3936
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x1364
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3952
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x1374
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3968
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x1384
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:3984
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x1394
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:4000
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x13a4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:4016
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x13b4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:4032
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x13c4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:4048
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x13d4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:4064
+; GFX9-FLATSCR-NEXT: s_movk_i32 s11, 0x13e4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s11 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:4080
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x13e4
+; GFX9-FLATSCR-NEXT: ;;#ASMSTART
+; GFX9-FLATSCR-NEXT: ;;#ASMEND
+; GFX9-FLATSCR-NEXT: ;;#ASMSTART
+; GFX9-FLATSCR-NEXT: ;;#ASMEND
+; GFX9-FLATSCR-NEXT: ;;#ASMSTART
+; GFX9-FLATSCR-NEXT: ;;#ASMEND
+; GFX9-FLATSCR-NEXT: ;;#ASMSTART
+; GFX9-FLATSCR-NEXT: ;;#ASMEND
+; GFX9-FLATSCR-NEXT: ;;#ASMSTART
+; GFX9-FLATSCR-NEXT: ;;#ASMEND
+; GFX9-FLATSCR-NEXT: ;;#ASMSTART
+; GFX9-FLATSCR-NEXT: ;;#ASMEND
+; GFX9-FLATSCR-NEXT: ;;#ASMSTART
+; GFX9-FLATSCR-NEXT: ;;#ASMEND
+; GFX9-FLATSCR-NEXT: v_add_co_u32_e32 v4, vcc, s0, v5
+; GFX9-FLATSCR-NEXT: v_addc_co_u32_e32 v6, vcc, 0, v6, vcc
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:4080
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x13d4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:4064
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x13c4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:4048
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x13b4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:4032
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x13a4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:4016
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x1394
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:4000
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x1384
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3984
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x1374
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3968
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x1364
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3952
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x1354
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3936
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x1344
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3920
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x1334
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3904
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x1324
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3888
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x1314
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3872
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x1304
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3856
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x12f4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3840
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x12e4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3824
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x12d4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3808
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x12c4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3792
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x12b4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3776
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x12a4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3760
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x1294
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3744
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x1284
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3728
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x1274
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3712
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x1264
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3696
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x1254
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3680
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x1244
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3664
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x1234
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3648
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x1224
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3632
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x1214
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3616
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x1204
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3600
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x11f4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3584
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x11e4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3568
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x11d4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3552
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x11c4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3536
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x11b4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3520
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x11a4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3504
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x1194
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3488
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x1184
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3472
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x1174
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3456
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x1164
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3440
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x1154
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3424
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x1144
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3408
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x1134
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3392
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x1124
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3376
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x1114
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3360
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x1104
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3344
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x10f4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3328
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x10e4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3312
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x10d4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3296
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x10c4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3280
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x10b4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3264
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x10a4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3248
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x1094
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3232
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x1084
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3216
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x1074
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3200
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x1064
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3184
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x1054
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3168
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x1044
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3152
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x1034
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3136
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x1024
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3120
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x1014
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3104
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x1004
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3088
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xff4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3072
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xfe4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3056
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xfd4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3040
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xfc4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3024
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xfb4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:3008
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xfa4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2992
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xf94
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2976
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xf84
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2960
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xf74
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2944
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xf64
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2928
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xf54
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2912
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xf44
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2896
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xf34
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2880
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xf24
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2864
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xf14
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2848
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xf04
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2832
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xef4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2816
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xee4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2800
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xed4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2784
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xec4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2768
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xeb4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2752
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xea4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2736
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xe94
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2720
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xe84
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2704
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xe74
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2688
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xe64
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2672
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xe54
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2656
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xe44
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2640
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xe34
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2624
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xe24
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2608
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xe14
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2592
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xe04
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2576
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xdf4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2560
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xde4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2544
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xdd4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2528
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xdc4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2512
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xdb4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2496
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xda4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2480
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xd94
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2464
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xd84
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2448
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xd74
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2432
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xd64
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2416
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xd54
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2400
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xd44
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2384
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xd34
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2368
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xd24
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2352
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xd14
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2336
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xd04
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2320
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xcf4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2304
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xce4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2288
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xcd4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2272
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xcc4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2256
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xcb4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2240
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xca4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2224
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xc94
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2208
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xc84
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2192
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xc74
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2176
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xc64
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2160
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xc54
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2144
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xc44
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2128
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xc34
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2112
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xc24
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2096
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xc14
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2080
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xc04
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2064
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xbf4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2048
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xbe4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2032
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xbd4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2016
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xbc4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2000
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xbb4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1984
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xba4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1968
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xb94
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1952
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xb84
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1936
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xb74
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1920
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xb64
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1904
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xb54
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1888
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xb44
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1872
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xb34
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1856
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xb24
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1840
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xb14
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1824
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xb04
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1808
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xaf4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1792
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xae4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1776
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xad4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1760
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xac4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1744
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xab4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1728
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xaa4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1712
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xa94
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1696
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xa84
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1680
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xa74
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1664
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xa64
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1648
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xa54
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1632
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xa44
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1616
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xa34
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1600
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xa24
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1584
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xa14
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1568
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0xa04
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1552
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x9f4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1536
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x9e4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1520
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x9d4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1504
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x9c4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1488
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x9b4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1472
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x9a4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1456
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x994
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1440
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x984
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1424
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x974
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1408
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x964
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1392
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x954
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1376
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x944
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1360
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x934
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1344
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x924
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1328
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x914
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1312
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x904
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1296
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x8f4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1280
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x8e4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1264
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x8d4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1248
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x8c4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1232
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x8b4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1216
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x8a4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1200
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x894
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1184
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x884
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1168
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x874
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1152
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x864
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1136
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x854
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1120
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x844
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1104
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x834
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1088
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x824
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1072
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x814
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1056
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x804
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1040
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x7f4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1024
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x7e4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1008
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x7d4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:992
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x7c4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:976
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x7b4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:960
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x7a4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:944
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x794
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:928
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x784
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:912
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x774
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:896
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x764
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:880
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x754
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:864
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x744
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:848
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x734
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:832
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x724
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:816
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x714
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:800
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x704
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:784
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x6f4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:768
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x6e4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:752
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x6d4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:736
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x6c4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:720
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x6b4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:704
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x6a4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:688
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x694
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:672
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x684
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:656
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x674
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:640
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x664
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:624
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x654
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:608
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x644
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:592
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x634
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:576
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x624
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:560
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x614
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:544
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x604
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:528
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x5f4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:512
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x5e4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:496
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x5d4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:480
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x5c4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:464
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x5b4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:448
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x5a4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:432
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x594
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:416
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x584
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:400
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x574
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:384
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x564
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:368
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x554
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:352
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x544
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:336
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x534
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:320
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x524
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:304
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x514
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:288
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x504
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:272
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x4f4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:256
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x4e4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:240
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x4d4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:224
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x4c4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:208
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x4b4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:192
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x4a4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:176
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x494
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:160
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x484
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:144
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x474
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:128
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x464
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:112
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x454
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:96
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x444
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:80
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x434
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:64
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x424
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:48
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x414
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:32
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s2, 0x404
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:16
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1]
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x3f4
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: v_add_co_u32_e32 v0, vcc, s10, v4
+; GFX9-FLATSCR-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v6, vcc
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x3e4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:4080
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x3d4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:4064
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x3c4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:4048
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x3b4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:4032
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x3a4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:4016
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x394
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:4000
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x384
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:3984
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x374
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:3968
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: v_add_co_u32_e32 v0, vcc, s9, v4
+; GFX9-FLATSCR-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v6, vcc
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x364
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:4080
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x354
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:4064
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x344
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:4048
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x334
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:4032
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x324
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:4016
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x314
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:4000
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x304
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:3984
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x2f4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:3968
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: v_add_co_u32_e32 v0, vcc, s8, v4
+; GFX9-FLATSCR-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v6, vcc
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x2e4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:4080
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x2d4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:4064
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x2c4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:4048
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x2b4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:4032
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x2a4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:4016
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x294
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:4000
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x284
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:3984
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x274
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:3968
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: v_add_co_u32_e32 v0, vcc, s7, v4
+; GFX9-FLATSCR-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v6, vcc
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x264
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:4080
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x254
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:4064
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x244
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:4048
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x234
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:4032
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x224
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:4016
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x214
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:4000
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x204
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:3984
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x1f4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:3968
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: v_add_co_u32_e32 v0, vcc, s6, v4
+; GFX9-FLATSCR-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v6, vcc
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x1e4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:4080
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x1d4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:4064
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x1c4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:4048
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x1b4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:4032
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x1a4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:4016
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x194
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:4000
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x184
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:3984
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x174
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:3968
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: v_add_co_u32_e32 v0, vcc, s5, v4
+; GFX9-FLATSCR-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v6, vcc
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x164
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:4080
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x154
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:4064
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x144
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:4048
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x134
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:4032
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x124
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:4016
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x114
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:4000
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x104
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:3984
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0xf4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:3968
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: v_add_co_u32_e32 v0, vcc, s4, v4
+; GFX9-FLATSCR-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v6, vcc
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0xe4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:4080
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0xd4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:4064
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0xc4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:4048
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0xb4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:4032
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0xa4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:4016
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x94
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:4000
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x84
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:3984
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x74
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[7:10], off offset:3968
+; GFX9-FLATSCR-NEXT: v_add_co_u32_e32 v0, vcc, 0x80, v4
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[2:5], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v6, vcc
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x64
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[2:5], off offset:4080
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[2:5], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x54
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[2:5], off offset:4064
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[2:5], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x44
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[2:5], off offset:4048
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[2:5], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_mov_b32 s0, 52
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[2:5], off offset:4032
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[2:5], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_mov_b32 s0, 36
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[2:5], off offset:4016
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[2:5], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_mov_b32 s0, 20
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[2:5], off offset:4000
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[2:5], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_mov_b32 s0, 4
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[2:5], off offset:3984
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[2:5], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[2:5], off offset:3968
+; GFX9-FLATSCR-NEXT: s_endpgm
+;
+; GFX10-FLATSCR-LABEL: test:
+; GFX10-FLATSCR: ; %bb.0: ; %entry
+; GFX10-FLATSCR-NEXT: s_add_u32 s2, s2, s5
+; GFX10-FLATSCR-NEXT: s_addc_u32 s3, s3, 0
+; GFX10-FLATSCR-NEXT: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s2
+; GFX10-FLATSCR-NEXT: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s3
+; GFX10-FLATSCR-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX10-FLATSCR-NEXT: v_mbcnt_lo_u32_b32 v0, -1, 0
+; GFX10-FLATSCR-NEXT: v_mbcnt_hi_u32_b32 v0, -1, v0
+; GFX10-FLATSCR-NEXT: v_lshlrev_b32_e32 v5, 13, v0
+; GFX10-FLATSCR-NEXT: s_waitcnt lgkmcnt(0)
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v4, s4, s2, v5
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e64 v22, null, s3, 0, s4
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x804
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v0, vcc_lo, 0x80, v4
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v22, vcc_lo
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v2, vcc_lo, 0x100, v4
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v3, vcc_lo, 0, v22, vcc_lo
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:1920
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], off offset:4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:1936
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], off offset:20 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:1952
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], off offset:36 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:1968
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], off offset:52 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:1984
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], off offset:68 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:2000
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], off offset:84 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:2016
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], off offset:100 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:2032
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], off offset:116 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[2:3], off offset:1920
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], off offset:132 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[2:3], off offset:1936
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], off offset:148 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[2:3], off offset:1952
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], off offset:164 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[2:3], off offset:1968
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], off offset:180 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[2:3], off offset:1984
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], off offset:196 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[2:3], off offset:2000
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], off offset:212 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[2:3], off offset:2016
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], off offset:228 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[2:3], off offset:2032
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], off offset:244 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v6, vcc_lo, 0x180, v4
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v7, vcc_lo, 0, v22, vcc_lo
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[8:11], v[6:7], off offset:1920
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[8:11], off offset:260 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[8:11], v[6:7], off offset:1936
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[8:11], off offset:276 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[8:11], v[6:7], off offset:1952
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[8:11], off offset:292 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[8:11], v[6:7], off offset:1968
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[8:11], off offset:308 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[8:11], v[6:7], off offset:1984
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[8:11], off offset:324 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[8:11], v[6:7], off offset:2000
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[8:11], off offset:340 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[8:11], v[6:7], off offset:2016
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[8:11], off offset:356 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[8:11], v[6:7], off offset:2032
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[8:11], off offset:372 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v8, vcc_lo, 0x200, v4
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v9, vcc_lo, 0, v22, vcc_lo
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[10:13], v[8:9], off offset:1920
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[10:13], off offset:388 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[10:13], v[8:9], off offset:1936
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[10:13], off offset:404 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[10:13], v[8:9], off offset:1952
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[10:13], off offset:420 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[10:13], v[8:9], off offset:1968
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[10:13], off offset:436 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[10:13], v[8:9], off offset:1984
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[10:13], off offset:452 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[10:13], v[8:9], off offset:2000
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[10:13], off offset:468 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[10:13], v[8:9], off offset:2016
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[10:13], off offset:484 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[10:13], v[8:9], off offset:2032
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[10:13], off offset:500 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v10, vcc_lo, 0x280, v4
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v11, vcc_lo, 0, v22, vcc_lo
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[12:15], v[10:11], off offset:1920
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[12:15], off offset:516 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[12:15], v[10:11], off offset:1936
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[12:15], off offset:532 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[12:15], v[10:11], off offset:1952
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[12:15], off offset:548 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[12:15], v[10:11], off offset:1968
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[12:15], off offset:564 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[12:15], v[10:11], off offset:1984
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[12:15], off offset:580 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[12:15], v[10:11], off offset:2000
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[12:15], off offset:596 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[12:15], v[10:11], off offset:2016
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[12:15], off offset:612 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[12:15], v[10:11], off offset:2032
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[12:15], off offset:628 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v12, vcc_lo, 0x300, v4
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v13, vcc_lo, 0, v22, vcc_lo
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[14:17], v[12:13], off offset:1920
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[14:17], off offset:644 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[14:17], v[12:13], off offset:1936
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[14:17], off offset:660 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[14:17], v[12:13], off offset:1952
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[14:17], off offset:676 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[14:17], v[12:13], off offset:1968
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[14:17], off offset:692 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[14:17], v[12:13], off offset:1984
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[14:17], off offset:708 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[14:17], v[12:13], off offset:2000
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[14:17], off offset:724 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[14:17], v[12:13], off offset:2016
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[14:17], off offset:740 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[14:17], v[12:13], off offset:2032
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[14:17], off offset:756 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v14, vcc_lo, 0x380, v4
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v15, vcc_lo, 0, v22, vcc_lo
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[16:19], v[14:15], off offset:1920
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[16:19], off offset:772 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[16:19], v[14:15], off offset:1936
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[16:19], off offset:788 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[16:19], v[14:15], off offset:1952
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[16:19], off offset:804 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[16:19], v[14:15], off offset:1968
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[16:19], off offset:820 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[16:19], v[14:15], off offset:1984
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[16:19], off offset:836 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[16:19], v[14:15], off offset:2000
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[16:19], off offset:852 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[16:19], v[14:15], off offset:2016
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[16:19], off offset:868 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[16:19], v[14:15], off offset:2032
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[16:19], off offset:884 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v16, vcc_lo, 0x400, v4
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v17, vcc_lo, 0, v22, vcc_lo
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[18:21], v[16:17], off offset:1920
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[18:21], off offset:900 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[18:21], v[16:17], off offset:1936
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[18:21], off offset:916 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[18:21], v[16:17], off offset:1952
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[18:21], off offset:932 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[18:21], v[16:17], off offset:1968
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[18:21], off offset:948 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[18:21], v[16:17], off offset:1984
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[18:21], off offset:964 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[18:21], v[16:17], off offset:2000
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[18:21], off offset:980 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[18:21], v[16:17], off offset:2016
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[18:21], off offset:996 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[18:21], v[16:17], off offset:2032
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[18:21], off offset:1012 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v18, vcc_lo, 0x480, v4
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v19, vcc_lo, 0, v22, vcc_lo
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v20, vcc_lo, 0x500, v4
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v21, vcc_lo, 0, v22, vcc_lo
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[23:26], v[18:19], off offset:1920
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[23:26], off offset:1028 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[23:26], v[18:19], off offset:1936
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[23:26], off offset:1044 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[23:26], v[18:19], off offset:1952
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[23:26], off offset:1060 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[23:26], v[18:19], off offset:1968
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[23:26], off offset:1076 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[23:26], v[18:19], off offset:1984
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[23:26], off offset:1092 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[23:26], v[18:19], off offset:2000
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[23:26], off offset:1108 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[23:26], v[18:19], off offset:2016
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[23:26], off offset:1124 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[23:26], v[18:19], off offset:2032
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[23:26], off offset:1140 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[23:26], v[20:21], off offset:1920
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[23:26], off offset:1156 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[23:26], v[20:21], off offset:1936
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[23:26], off offset:1172 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[23:26], v[20:21], off offset:1952
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[23:26], off offset:1188 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[23:26], v[20:21], off offset:1968
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[23:26], off offset:1204 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[23:26], v[20:21], off offset:1984
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[23:26], off offset:1220 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[23:26], v[20:21], off offset:2000
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[23:26], off offset:1236 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[23:26], v[20:21], off offset:2016
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[23:26], off offset:1252 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[23:26], v[20:21], off offset:2032
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v20, vcc_lo, 0x580, v4
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v21, vcc_lo, 0, v22, vcc_lo
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[23:26], off offset:1268 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[23:26], v[20:21], off offset:1920
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[23:26], off offset:1284 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[23:26], v[20:21], off offset:1936
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[23:26], off offset:1300 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[23:26], v[20:21], off offset:1952
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[23:26], off offset:1316 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[23:26], v[20:21], off offset:1968
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[23:26], off offset:1332 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[23:26], v[20:21], off offset:1984
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[23:26], off offset:1348 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[23:26], v[20:21], off offset:2000
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[23:26], off offset:1364 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[23:26], v[20:21], off offset:2016
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[23:26], off offset:1380 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[23:26], v[20:21], off offset:2032
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v20, vcc_lo, 0x600, v4
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v21, vcc_lo, 0, v22, vcc_lo
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[23:26], off offset:1396 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[23:26], v[20:21], off offset:1920
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[23:26], off offset:1412 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[23:26], v[20:21], off offset:1936
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[23:26], off offset:1428 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[23:26], v[20:21], off offset:1952
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[23:26], off offset:1444 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[23:26], v[20:21], off offset:1968
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[23:26], off offset:1460 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[23:26], v[20:21], off offset:1984
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[23:26], off offset:1476 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[23:26], v[20:21], off offset:2000
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[23:26], off offset:1492 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[23:26], v[20:21], off offset:2016
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[23:26], off offset:1508 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[23:26], v[20:21], off offset:2032
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v20, vcc_lo, 0x680, v4
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v21, vcc_lo, 0, v22, vcc_lo
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[23:26], off offset:1524 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[23:26], v[20:21], off offset:1920
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[23:26], off offset:1540 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[23:26], v[20:21], off offset:1936
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[23:26], off offset:1556 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[23:26], v[20:21], off offset:1952
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[23:26], off offset:1572 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[23:26], v[20:21], off offset:1968
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[23:26], off offset:1588 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[23:26], v[20:21], off offset:1984
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[23:26], off offset:1604 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[23:26], v[20:21], off offset:2000
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[23:26], off offset:1620 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[23:26], v[20:21], off offset:2016
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[23:26], off offset:1636 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[23:26], v[20:21], off offset:2032
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v20, vcc_lo, 0x700, v4
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v21, vcc_lo, 0, v22, vcc_lo
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[23:26], off offset:1652 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[23:26], v[20:21], off offset:1920
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[23:26], off offset:1668 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[23:26], v[20:21], off offset:1936
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[23:26], off offset:1684 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[23:26], v[20:21], off offset:1952
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[23:26], off offset:1700 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[23:26], v[20:21], off offset:1968
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[23:26], off offset:1716 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[23:26], v[20:21], off offset:1984
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[23:26], off offset:1732 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[23:26], v[20:21], off offset:2000
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[23:26], off offset:1748 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[23:26], v[20:21], off offset:2016
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[23:26], off offset:1764 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[23:26], v[20:21], off offset:2032
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v20, vcc_lo, 0x780, v4
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v21, vcc_lo, 0, v22, vcc_lo
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v0, vcc_lo, 0x780, v0
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v1, vcc_lo
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[23:26], off offset:1780 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[22:25], v[20:21], off offset:1920
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[22:25], off offset:1796 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[22:25], v[20:21], off offset:1936
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[22:25], off offset:1812 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[22:25], v[20:21], off offset:1952
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[22:25], off offset:1828 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[22:25], v[20:21], off offset:1968
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[22:25], off offset:1844 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[22:25], v[20:21], off offset:1984
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[22:25], off offset:1860 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[22:25], v[20:21], off offset:2000
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[22:25], off offset:1876 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[22:25], v[20:21], off offset:2016
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[22:25], off offset:1892 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[20:23], v[20:21], off offset:2032
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[20:23], off offset:1908 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[20:23], v[0:1], off offset:1920
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[20:23], off offset:1924 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[20:23], v[0:1], off offset:1936
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[20:23], off offset:1940 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[20:23], v[0:1], off offset:1952
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[20:23], off offset:1956 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[20:23], v[0:1], off offset:1968
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[20:23], off offset:1972 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[20:23], v[0:1], off offset:1984
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[20:23], off offset:1988 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[20:23], v[0:1], off offset:2000
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[20:23], off offset:2004 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[20:23], v[0:1], off offset:2016
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[20:23], off offset:2020 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[20:23], v[0:1], off offset:2032
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v0, vcc_lo, 0x780, v2
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v3, vcc_lo
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[20:23], off offset:2036 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[20:23], v[0:1], off offset:1920
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[20:23], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[20:23], v[0:1], off offset:1936
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x814
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[20:23], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[20:23], v[0:1], off offset:1952
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x824
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[20:23], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[20:23], v[0:1], off offset:1968
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x834
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[20:23], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[20:23], v[0:1], off offset:1984
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x844
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[20:23], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[20:23], v[0:1], off offset:2000
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x854
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[20:23], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: s_clause 0x1
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[20:23], v[0:1], off offset:2016
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v[0:1], off offset:2032
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x864
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(1)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[20:23], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x874
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v0, vcc_lo, 0x780, v6
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v7, vcc_lo
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x884
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[20:23], v[0:1], off offset:1920
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[20:23], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[20:23], v[0:1], off offset:1936
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x894
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[20:23], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[20:23], v[0:1], off offset:1952
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x8a4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[20:23], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[20:23], v[0:1], off offset:1968
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x8b4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[20:23], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[20:23], v[0:1], off offset:1984
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x8c4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[20:23], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[20:23], v[0:1], off offset:2000
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x8d4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[20:23], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: s_clause 0x1
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[20:23], v[0:1], off offset:2016
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v[0:1], off offset:2032
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x8e4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(1)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[20:23], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x8f4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v0, vcc_lo, 0x780, v8
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v9, vcc_lo
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x904
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:1920
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:1936
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x914
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:1952
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x924
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:1968
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x934
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:1984
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x944
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:2000
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x954
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: s_clause 0x1
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:2016
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v[0:1], off offset:2032
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x964
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(1)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x974
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v0, vcc_lo, 0x780, v10
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v11, vcc_lo
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x984
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:1920
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:1936
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x994
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:1952
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x9a4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:1968
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x9b4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:1984
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x9c4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:2000
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x9d4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: s_clause 0x1
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:2016
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v[0:1], off offset:2032
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x9e4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(1)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x9f4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v0, vcc_lo, 0x780, v12
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v13, vcc_lo
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xa04
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:1920
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:1936
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xa14
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:1952
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xa24
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:1968
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xa34
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:1984
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xa44
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:2000
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xa54
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: s_clause 0x1
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:2016
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v[0:1], off offset:2032
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xa64
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(1)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xa74
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v0, vcc_lo, 0x780, v14
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v15, vcc_lo
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xa84
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:1920
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:1936
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xa94
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:1952
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xaa4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:1968
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xab4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:1984
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xac4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:2000
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xad4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: s_clause 0x1
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:2016
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v[0:1], off offset:2032
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xae4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(1)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xaf4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v0, vcc_lo, 0x780, v16
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v17, vcc_lo
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xb04
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:1920
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:1936
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xb14
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:1952
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xb24
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:1968
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xb34
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:1984
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xb44
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:2000
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xb54
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: s_clause 0x1
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:2016
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v[0:1], off offset:2032
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xb64
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(1)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xb74
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v0, vcc_lo, 0x780, v18
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v19, vcc_lo
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xb84
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:1920
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:1936
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xb94
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:1952
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xba4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:1968
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xbb4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:1984
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xbc4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:2000
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xbd4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: s_clause 0x1
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[6:9], v[0:1], off offset:2016
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v[0:1], off offset:2032
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xbe4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(1)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[6:9], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xbf4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3]
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xc04
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:16
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xc14
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:32
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xc24
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:48
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xc34
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:64
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xc44
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:80
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xc54
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:96
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xc64
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:112
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xc74
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:128
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xc84
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:144
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xc94
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:160
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xca4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:176
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xcb4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:192
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xcc4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:208
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xcd4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:224
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xce4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:240
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xcf4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:256
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xd04
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:272
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xd14
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:288
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xd24
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:304
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xd34
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:320
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xd44
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:336
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xd54
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:352
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xd64
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:368
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xd74
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:384
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xd84
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:400
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xd94
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:416
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xda4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:432
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xdb4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:448
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xdc4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:464
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xdd4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:480
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xde4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:496
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xdf4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:512
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xe04
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:528
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xe14
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:544
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xe24
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:560
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xe34
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:576
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xe44
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:592
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xe54
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:608
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xe64
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:624
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xe74
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:640
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xe84
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:656
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xe94
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:672
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xea4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:688
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xeb4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:704
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xec4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:720
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xed4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:736
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xee4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:752
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xef4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:768
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xf04
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:784
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xf14
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:800
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xf24
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:816
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xf34
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:832
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xf44
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:848
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xf54
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:864
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xf64
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:880
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xf74
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:896
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xf84
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:912
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xf94
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:928
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xfa4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:944
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xfb4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:960
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xfc4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:976
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xfd4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:992
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xfe4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1008
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0xff4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1024
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x1004
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1040
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x1014
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1056
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x1024
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1072
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x1034
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1088
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x1044
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1104
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x1054
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1120
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x1064
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1136
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x1074
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1152
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x1084
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1168
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x1094
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1184
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x10a4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1200
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x10b4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1216
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x10c4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1232
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x10d4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1248
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x10e4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1264
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x10f4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1280
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x1104
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1296
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x1114
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1312
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x1124
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1328
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x1134
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1344
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x1144
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1360
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x1154
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1376
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x1164
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1392
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x1174
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1408
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x1184
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1424
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x1194
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1440
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x11a4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1456
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x11b4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1472
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x11c4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1488
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x11d4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1504
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x11e4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1520
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x11f4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1536
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x1204
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1552
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x1214
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1568
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x1224
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1584
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x1234
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1600
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x1244
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1616
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x1254
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1632
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x1264
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1648
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x1274
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1664
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x1284
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1680
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x1294
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1696
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x12a4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1712
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x12b4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1728
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x12c4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1744
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x12d4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1760
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x12e4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1776
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x12f4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1792
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x1304
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1808
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x1314
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1824
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x1324
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1840
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x1334
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1856
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x1344
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1872
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x1354
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1888
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x1364
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1904
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x1374
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1920
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x1384
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1936
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x1394
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1952
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x13a4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1968
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x13b4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:1984
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x13c4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2000
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x13d4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2016
+; GFX10-FLATSCR-NEXT: s_movk_i32 s4, 0x13e4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s4 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v5, s[2:3] offset:2032
+; GFX10-FLATSCR-NEXT: ;;#ASMSTART
+; GFX10-FLATSCR-NEXT: ;;#ASMEND
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v4, s2, s0, v5
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e64 v6, null, s1, 0, s2
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x13e4
+; GFX10-FLATSCR-NEXT: ;;#ASMSTART
+; GFX10-FLATSCR-NEXT: ;;#ASMEND
+; GFX10-FLATSCR-NEXT: ;;#ASMSTART
+; GFX10-FLATSCR-NEXT: ;;#ASMEND
+; GFX10-FLATSCR-NEXT: ;;#ASMSTART
+; GFX10-FLATSCR-NEXT: ;;#ASMEND
+; GFX10-FLATSCR-NEXT: ;;#ASMSTART
+; GFX10-FLATSCR-NEXT: ;;#ASMEND
+; GFX10-FLATSCR-NEXT: ;;#ASMSTART
+; GFX10-FLATSCR-NEXT: ;;#ASMEND
+; GFX10-FLATSCR-NEXT: ;;#ASMSTART
+; GFX10-FLATSCR-NEXT: ;;#ASMEND
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2032
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x13d4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2016
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x13c4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:2000
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x13b4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1984
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x13a4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1968
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x1394
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1952
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x1384
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1936
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x1374
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1920
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x1364
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1904
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x1354
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1888
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x1344
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1872
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x1334
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1856
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x1324
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1840
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x1314
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1824
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x1304
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1808
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x12f4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1792
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x12e4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1776
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x12d4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1760
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x12c4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1744
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x12b4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1728
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x12a4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1712
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x1294
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1696
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x1284
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1680
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x1274
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1664
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x1264
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1648
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x1254
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1632
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x1244
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1616
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x1234
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1600
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x1224
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1584
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x1214
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1568
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x1204
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1552
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x11f4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1536
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x11e4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1520
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x11d4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1504
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x11c4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1488
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x11b4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1472
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x11a4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1456
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x1194
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1440
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x1184
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1424
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x1174
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1408
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x1164
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1392
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x1154
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1376
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x1144
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1360
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x1134
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1344
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x1124
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1328
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x1114
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1312
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x1104
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1296
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x10f4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1280
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x10e4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1264
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x10d4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1248
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x10c4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1232
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x10b4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1216
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x10a4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1200
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x1094
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1184
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x1084
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1168
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x1074
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1152
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x1064
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1136
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x1054
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1120
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x1044
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1104
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x1034
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1088
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x1024
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1072
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x1014
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1056
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0x1004
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1040
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xff4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1024
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xfe4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:1008
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xfd4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:992
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xfc4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:976
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xfb4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:960
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xfa4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:944
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xf94
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:928
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xf84
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:912
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xf74
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:896
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xf64
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:880
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xf54
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:864
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xf44
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:848
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xf34
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:832
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xf24
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:816
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xf14
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:800
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xf04
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:784
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xef4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:768
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xee4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:752
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xed4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:736
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xec4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:720
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xeb4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:704
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xea4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:688
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xe94
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:672
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xe84
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:656
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xe74
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:640
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xe64
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:624
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xe54
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:608
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xe44
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:592
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xe34
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:576
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xe24
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:560
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xe14
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:544
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xe04
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:528
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xdf4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:512
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xde4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:496
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xdd4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:480
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xdc4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:464
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xdb4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:448
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xda4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:432
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xd94
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:416
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xd84
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:400
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xd74
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:384
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xd64
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:368
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xd54
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:352
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xd44
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:336
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xd34
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:320
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xd24
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:304
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xd14
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:288
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xd04
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:272
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xcf4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:256
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xce4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:240
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xcd4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:224
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xcc4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:208
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xcb4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:192
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xca4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:176
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xc94
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:160
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xc84
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:144
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xc74
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:128
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xc64
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:112
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xc54
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:96
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xc44
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:80
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xc34
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:64
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xc24
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:48
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xc14
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:32
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s2, 0xc04
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1] offset:16
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s2 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v5, v[0:3], s[0:1]
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0xbf4
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v0, vcc_lo, 0x480, v4
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v1, vcc_lo, 0, v6, vcc_lo
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v2, vcc_lo, 0x780, v0
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0xbe4
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v3, vcc_lo, 0, v1, vcc_lo
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[2:3], v[7:10], off offset:2032
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0xbd4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[2:3], v[7:10], off offset:2016
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0xbc4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[2:3], v[7:10], off offset:2000
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0xbb4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[2:3], v[7:10], off offset:1984
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0xba4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[2:3], v[7:10], off offset:1968
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0xb94
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[2:3], v[7:10], off offset:1952
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0xb84
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[2:3], v[7:10], off offset:1936
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[7:10], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0xb74
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[2:3], v[7:10], off offset:1920
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[9:12], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v2, vcc_lo, 0x400, v4
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v3, vcc_lo, 0, v6, vcc_lo
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0xb64
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v7, vcc_lo, 0x780, v2
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v8, vcc_lo, 0, v3, vcc_lo
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[7:8], v[9:12], off offset:2032
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[9:12], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0xb54
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[7:8], v[9:12], off offset:2016
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[9:12], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0xb44
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[7:8], v[9:12], off offset:2000
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[9:12], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0xb34
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[7:8], v[9:12], off offset:1984
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[9:12], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0xb24
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[7:8], v[9:12], off offset:1968
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[9:12], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0xb14
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[7:8], v[9:12], off offset:1952
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[9:12], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0xb04
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[7:8], v[9:12], off offset:1936
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[9:12], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0xaf4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[7:8], v[9:12], off offset:1920
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[11:14], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v7, vcc_lo, 0x380, v4
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v8, vcc_lo, 0, v6, vcc_lo
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0xae4
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v9, vcc_lo, 0x780, v7
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v10, vcc_lo, 0, v8, vcc_lo
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[9:10], v[11:14], off offset:2032
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[11:14], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0xad4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[9:10], v[11:14], off offset:2016
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[11:14], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0xac4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[9:10], v[11:14], off offset:2000
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[11:14], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0xab4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[9:10], v[11:14], off offset:1984
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[11:14], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0xaa4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[9:10], v[11:14], off offset:1968
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[11:14], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0xa94
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[9:10], v[11:14], off offset:1952
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[11:14], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0xa84
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[9:10], v[11:14], off offset:1936
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[11:14], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0xa74
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[9:10], v[11:14], off offset:1920
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[13:16], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v9, vcc_lo, 0x300, v4
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v10, vcc_lo, 0, v6, vcc_lo
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0xa64
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v11, vcc_lo, 0x780, v9
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v12, vcc_lo, 0, v10, vcc_lo
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[11:12], v[13:16], off offset:2032
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[13:16], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0xa54
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[11:12], v[13:16], off offset:2016
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[13:16], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0xa44
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[11:12], v[13:16], off offset:2000
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[13:16], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0xa34
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[11:12], v[13:16], off offset:1984
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[13:16], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0xa24
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[11:12], v[13:16], off offset:1968
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[13:16], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0xa14
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[11:12], v[13:16], off offset:1952
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[13:16], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0xa04
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[11:12], v[13:16], off offset:1936
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[13:16], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0x9f4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[11:12], v[13:16], off offset:1920
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[15:18], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v11, vcc_lo, 0x280, v4
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v12, vcc_lo, 0, v6, vcc_lo
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0x9e4
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v13, vcc_lo, 0x780, v11
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v14, vcc_lo, 0, v12, vcc_lo
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[13:14], v[15:18], off offset:2032
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[15:18], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0x9d4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[13:14], v[15:18], off offset:2016
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[15:18], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0x9c4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[13:14], v[15:18], off offset:2000
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[15:18], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0x9b4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[13:14], v[15:18], off offset:1984
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[15:18], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0x9a4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[13:14], v[15:18], off offset:1968
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[15:18], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0x994
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[13:14], v[15:18], off offset:1952
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[15:18], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0x984
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[13:14], v[15:18], off offset:1936
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[15:18], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0x974
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[13:14], v[15:18], off offset:1920
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[17:20], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v13, vcc_lo, 0x200, v4
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v14, vcc_lo, 0, v6, vcc_lo
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0x964
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v15, vcc_lo, 0x780, v13
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v16, vcc_lo, 0, v14, vcc_lo
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[15:16], v[17:20], off offset:2032
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[17:20], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0x954
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[15:16], v[17:20], off offset:2016
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[17:20], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0x944
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[15:16], v[17:20], off offset:2000
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[17:20], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0x934
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[15:16], v[17:20], off offset:1984
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[17:20], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0x924
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[15:16], v[17:20], off offset:1968
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[17:20], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0x914
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[15:16], v[17:20], off offset:1952
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[17:20], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0x904
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[15:16], v[17:20], off offset:1936
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[17:20], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0x8f4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[15:16], v[17:20], off offset:1920
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[19:22], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v15, vcc_lo, 0x180, v4
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v16, vcc_lo, 0, v6, vcc_lo
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0x8e4
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v17, vcc_lo, 0x780, v15
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v18, vcc_lo, 0, v16, vcc_lo
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[17:18], v[19:22], off offset:2032
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[19:22], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0x8d4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[17:18], v[19:22], off offset:2016
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[19:22], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0x8c4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[17:18], v[19:22], off offset:2000
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[19:22], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0x8b4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[17:18], v[19:22], off offset:1984
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[19:22], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0x8a4
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[17:18], v[19:22], off offset:1968
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[19:22], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0x894
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[17:18], v[19:22], off offset:1952
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[19:22], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0x884
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[17:18], v[19:22], off offset:1936
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[19:22], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0x874
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[17:18], v[19:22], off offset:1920
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[21:24], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v17, vcc_lo, 0x100, v4
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v18, vcc_lo, 0, v6, vcc_lo
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0x864
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v19, vcc_lo, 0x780, v17
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v20, vcc_lo, 0, v18, vcc_lo
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[19:20], v[21:24], off offset:2032
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[21:24], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0x854
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[19:20], v[21:24], off offset:2016
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[21:24], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0x844
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[19:20], v[21:24], off offset:2000
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[21:24], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0x834
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[19:20], v[21:24], off offset:1984
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[21:24], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0x824
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[19:20], v[21:24], off offset:1968
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[21:24], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0x814
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[19:20], v[21:24], off offset:1952
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[21:24], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0x804
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[19:20], v[21:24], off offset:1936
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[21:24], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[19:20], v[21:24], off offset:1920
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[23:26], off, off offset:2036 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v19, vcc_lo, 0x80, v4
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v20, vcc_lo, 0, v6, vcc_lo
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v21, vcc_lo, 0x780, v19
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v22, vcc_lo, 0, v20, vcc_lo
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[21:22], v[23:26], off offset:2032
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[23:26], off, off offset:2020 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[21:22], v[23:26], off offset:2016
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[23:26], off, off offset:2004 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[21:22], v[23:26], off offset:2000
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[23:26], off, off offset:1988 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[21:22], v[23:26], off offset:1984
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[23:26], off, off offset:1972 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[21:22], v[23:26], off offset:1968
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[23:26], off, off offset:1956 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[21:22], v[23:26], off offset:1952
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[23:26], off, off offset:1940 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[21:22], v[23:26], off offset:1936
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[23:26], off, off offset:1924 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[21:22], v[23:26], off offset:1920
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[23:26], off, off offset:1908 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v21, vcc_lo, 0x780, v4
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v22, vcc_lo, 0, v6, vcc_lo
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[21:22], v[23:26], off offset:2032
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[23:26], off, off offset:1892 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[21:22], v[23:26], off offset:2016
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[23:26], off, off offset:1876 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[21:22], v[23:26], off offset:2000
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[23:26], off, off offset:1860 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[21:22], v[23:26], off offset:1984
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[23:26], off, off offset:1844 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[21:22], v[23:26], off offset:1968
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[23:26], off, off offset:1828 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[21:22], v[23:26], off offset:1952
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[23:26], off, off offset:1812 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[21:22], v[23:26], off offset:1936
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[23:26], off, off offset:1796 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[21:22], v[23:26], off offset:1920
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[23:26], off, off offset:1780 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v21, vcc_lo, 0x700, v4
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v22, vcc_lo, 0, v6, vcc_lo
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[21:22], v[23:26], off offset:2032
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[23:26], off, off offset:1764 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[21:22], v[23:26], off offset:2016
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[23:26], off, off offset:1748 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[21:22], v[23:26], off offset:2000
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[23:26], off, off offset:1732 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[21:22], v[23:26], off offset:1984
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[23:26], off, off offset:1716 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[21:22], v[23:26], off offset:1968
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[23:26], off, off offset:1700 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[21:22], v[23:26], off offset:1952
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[23:26], off, off offset:1684 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[21:22], v[23:26], off offset:1936
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[23:26], off, off offset:1668 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[21:22], v[23:26], off offset:1920
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[23:26], off, off offset:1652 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v21, vcc_lo, 0x680, v4
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v22, vcc_lo, 0, v6, vcc_lo
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[21:22], v[23:26], off offset:2032
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[23:26], off, off offset:1636 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[21:22], v[23:26], off offset:2016
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[23:26], off, off offset:1620 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[21:22], v[23:26], off offset:2000
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[23:26], off, off offset:1604 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[21:22], v[23:26], off offset:1984
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[23:26], off, off offset:1588 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[21:22], v[23:26], off offset:1968
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[23:26], off, off offset:1572 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[21:22], v[23:26], off offset:1952
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[23:26], off, off offset:1556 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[21:22], v[23:26], off offset:1936
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[23:26], off, off offset:1540 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[21:22], v[23:26], off offset:1920
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[23:26], off, off offset:1524 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v21, vcc_lo, 0x600, v4
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v22, vcc_lo, 0, v6, vcc_lo
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[21:22], v[23:26], off offset:2032
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[23:26], off, off offset:1508 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[21:22], v[23:26], off offset:2016
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[23:26], off, off offset:1492 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[21:22], v[23:26], off offset:2000
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[23:26], off, off offset:1476 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[21:22], v[23:26], off offset:1984
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[23:26], off, off offset:1460 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[21:22], v[23:26], off offset:1968
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[23:26], off, off offset:1444 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[21:22], v[23:26], off offset:1952
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[23:26], off, off offset:1428 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[21:22], v[23:26], off offset:1936
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[23:26], off, off offset:1412 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[21:22], v[23:26], off offset:1920
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[23:26], off, off offset:1396 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v21, vcc_lo, 0x580, v4
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v22, vcc_lo, 0, v6, vcc_lo
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v4, vcc_lo, 0x500, v4
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v5, vcc_lo, 0, v6, vcc_lo
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[21:22], v[23:26], off offset:2032
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[23:26], off, off offset:1380 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[21:22], v[23:26], off offset:2016
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[23:26], off, off offset:1364 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[21:22], v[23:26], off offset:2000
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[23:26], off, off offset:1348 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[21:22], v[23:26], off offset:1984
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[23:26], off, off offset:1332 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[21:22], v[23:26], off offset:1968
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[23:26], off, off offset:1316 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[21:22], v[23:26], off offset:1952
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[23:26], off, off offset:1300 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[21:22], v[23:26], off offset:1936
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[23:26], off, off offset:1284 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[21:22], v[23:26], off offset:1920
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[21:24], off, off offset:1268 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[4:5], v[21:24], off offset:2032
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[21:24], off, off offset:1252 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[4:5], v[21:24], off offset:2016
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[21:24], off, off offset:1236 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[4:5], v[21:24], off offset:2000
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[21:24], off, off offset:1220 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[4:5], v[21:24], off offset:1984
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[21:24], off, off offset:1204 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[4:5], v[21:24], off offset:1968
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[21:24], off, off offset:1188 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[4:5], v[21:24], off offset:1952
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[21:24], off, off offset:1172 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[4:5], v[21:24], off offset:1936
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[21:24], off, off offset:1156 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[4:5], v[21:24], off offset:1920
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[21:24], off, off offset:1140 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[21:24], off offset:2032
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[21:24], off, off offset:1124 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[21:24], off offset:2016
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[21:24], off, off offset:1108 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[21:24], off offset:2000
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[21:24], off, off offset:1092 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[21:24], off offset:1984
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[21:24], off, off offset:1076 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[21:24], off offset:1968
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[21:24], off, off offset:1060 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[21:24], off offset:1952
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[21:24], off, off offset:1044 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[21:24], off offset:1936
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[21:24], off, off offset:1028 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[0:1], v[21:24], off offset:1920
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[21:24], off, off offset:1012 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[2:3], v[21:24], off offset:2032
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[21:24], off, off offset:996 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[2:3], v[21:24], off offset:2016
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[21:24], off, off offset:980 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[2:3], v[21:24], off offset:2000
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[21:24], off, off offset:964 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[2:3], v[21:24], off offset:1984
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[21:24], off, off offset:948 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[2:3], v[21:24], off offset:1968
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[21:24], off, off offset:932 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[2:3], v[21:24], off offset:1952
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[21:24], off, off offset:916 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[2:3], v[21:24], off offset:1936
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[21:24], off, off offset:900 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[2:3], v[21:24], off offset:1920
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:884 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[7:8], v[0:3], off offset:2032
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:868 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[7:8], v[0:3], off offset:2016
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:852 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[7:8], v[0:3], off offset:2000
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:836 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[7:8], v[0:3], off offset:1984
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:820 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[7:8], v[0:3], off offset:1968
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:804 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[7:8], v[0:3], off offset:1952
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:788 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[7:8], v[0:3], off offset:1936
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:772 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[7:8], v[0:3], off offset:1920
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:756 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[9:10], v[0:3], off offset:2032
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:740 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[9:10], v[0:3], off offset:2016
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:724 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[9:10], v[0:3], off offset:2000
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:708 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[9:10], v[0:3], off offset:1984
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:692 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[9:10], v[0:3], off offset:1968
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:676 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[9:10], v[0:3], off offset:1952
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:660 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[9:10], v[0:3], off offset:1936
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:644 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[9:10], v[0:3], off offset:1920
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:628 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[11:12], v[0:3], off offset:2032
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:612 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[11:12], v[0:3], off offset:2016
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:596 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[11:12], v[0:3], off offset:2000
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:580 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[11:12], v[0:3], off offset:1984
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:564 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[11:12], v[0:3], off offset:1968
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:548 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[11:12], v[0:3], off offset:1952
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:532 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[11:12], v[0:3], off offset:1936
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:516 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[11:12], v[0:3], off offset:1920
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:500 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[13:14], v[0:3], off offset:2032
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:484 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[13:14], v[0:3], off offset:2016
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:468 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[13:14], v[0:3], off offset:2000
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:452 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[13:14], v[0:3], off offset:1984
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:436 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[13:14], v[0:3], off offset:1968
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:420 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[13:14], v[0:3], off offset:1952
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:404 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[13:14], v[0:3], off offset:1936
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:388 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[13:14], v[0:3], off offset:1920
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:372 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[15:16], v[0:3], off offset:2032
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:356 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[15:16], v[0:3], off offset:2016
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:340 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[15:16], v[0:3], off offset:2000
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:324 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[15:16], v[0:3], off offset:1984
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:308 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[15:16], v[0:3], off offset:1968
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:292 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[15:16], v[0:3], off offset:1952
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:276 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[15:16], v[0:3], off offset:1936
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:260 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[15:16], v[0:3], off offset:1920
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:244 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[17:18], v[0:3], off offset:2032
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:228 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[17:18], v[0:3], off offset:2016
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:212 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[17:18], v[0:3], off offset:2000
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:196 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[17:18], v[0:3], off offset:1984
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:180 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[17:18], v[0:3], off offset:1968
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:164 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[17:18], v[0:3], off offset:1952
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:148 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[17:18], v[0:3], off offset:1936
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:132 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[17:18], v[0:3], off offset:1920
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:116 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[19:20], v[0:3], off offset:2032
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:100 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[19:20], v[0:3], off offset:2016
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:84 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[19:20], v[0:3], off offset:2000
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:68 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[19:20], v[0:3], off offset:1984
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:52 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[19:20], v[0:3], off offset:1968
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:36 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[19:20], v[0:3], off offset:1952
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:20 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[19:20], v[0:3], off offset:1936
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, off offset:4 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[19:20], v[0:3], off offset:1920
+; GFX10-FLATSCR-NEXT: s_endpgm
entry:
%lo = call i32 @llvm.amdgcn.mbcnt.lo(i32 -1, i32 0)
%tid = call i32 @llvm.amdgcn.mbcnt.hi(i32 -1, i32 %lo)
ret void
}
-; CHECK-LABEL: test_limited_sgpr
-; GFX6: %bb.1:
-; GFX6: s_mov_b64 exec, 0xff
-; GFX6: buffer_store_dword [[SPILL_REG_0:v[0-9]+]]
-; GFX6-COUNT-8: v_writelane_b32 [[SPILL_REG_0]]
-; GFX6: v_mov_b32_e32 [[OFFSET_REG0:v[0-9]+]], 0x[[OFFSET0:[0-9a-f]+]]
-; GFX6: buffer_store_dword [[SPILL_REG_0]], [[OFFSET_REG0]], s{{\[[0-9]+:[0-9]+\]}}, 0 offen
-; GFX6: buffer_load_dword [[SPILL_REG_0]], off, s{{\[[0-9]+:[0-9]+\]}}, 0{{$}}
-; GFX6: s_mov_b64 exec, s
-
-
-; GFX6: s_mov_b64 exec, 0xff
-; GFX6: v_mov_b32_e32 [[RELOAD_OFFSET_REG0:v[0-9]+]], 0x[[RELOAD_OFFSET0:[0-9a-f]+]]
-; GFX6: buffer_store_dword [[RELOAD_REG_0:v[0-9]+]], off,
-; GFX6: buffer_load_dword [[RELOAD_REG_0]], [[RELOAD_OFFSET_REG0]], s{{\[[0-9]+:[0-9]+\]}}, 0 offen
-; GFX6-COUNT-8: v_readlane_b32 s{{[0-9]+}}, [[RELOAD_REG_0]]
-; GFX6: buffer_load_dword [[RELOAD_REG_0]], off, s{{\[[0-9]+:[0-9]+\]}}, 0{{$}}
-; GFX6: s_mov_b64 exec,
-
-
-; GFX6: s_mov_b64 exec, 0xff
-; GFX6: buffer_store_dword [[SPILL_REG_1:v[0-9]+]]
-; GFX6-COUNT-8: v_writelane_b32 [[SPILL_REG_1]]
-; GFX6: v_mov_b32_e32 [[OFFSET_REG1:v[0-9]+]], 0x[[OFFSET1:[0-9a-f]+]]
-; GFX6: buffer_store_dword [[SPILL_REG_1]], [[OFFSET_REG1]], s{{\[[0-9]+:[0-9]+\]}}, 0 offen
-; GFX6: buffer_load_dword [[SPILL_REG_1]], off, s{{\[[0-9]+:[0-9]+\]}}, 0{{$}}
-; GFX6: s_mov_b64 exec, s
-
-
-; GFX6: s_mov_b64 exec, 0xff
-; GFX6: v_mov_b32_e32 [[RELOAD_OFFSET_REG1:v[0-9]+]], 0x[[RELOAD_OFFSET1:[0-9a-f]+]]
-; GFX6: buffer_store_dword [[RELOAD_REG_1:v[0-9]+]], off,
-; GFX6: buffer_load_dword [[RELOAD_REG_1]], [[RELOAD_OFFSET_REG1]], s{{\[[0-9]+:[0-9]+\]}}, 0 offen
-; GFX6-COUNT-8: v_readlane_b32 s{{[0-9]+}}, [[RELOAD_REG_1]]
-; GFX6: buffer_load_dword [[RELOAD_REG_1]], off, s{{\[[0-9]+:[0-9]+\]}}, 0{{$}}
-; GFX6: s_mov_b64 exec,
-
-
-; GFX6: s_mov_b64 exec, 0xff
-; GFX6: buffer_store_dword [[SPILL_REG_2:v[0-9]+]]
-; GFX6-COUNT-8: v_writelane_b32 [[SPILL_REG_2]]
-; GFX6: v_mov_b32_e32 [[OFFSET_REG2:v[0-9]+]], 0x[[OFFSET2:[0-9a-f]+]]
-; GFX6: buffer_store_dword [[SPILL_REG_2]], [[OFFSET_REG2]], s{{\[[0-9]+:[0-9]+\]}}, 0 offen
-; GFX6: buffer_load_dword [[SPILL_REG_2]], off, s{{\[[0-9]+:[0-9]+\]}}, 0{{$}}
-; GFX6: s_mov_b64 exec, s
-
-
-; GFX6: s_mov_b64 exec, 0xff
-; GFX6: buffer_store_dword [[SPILL_REG_3:v[0-9]+]]
-; GFX6-COUNT-8: v_writelane_b32 [[SPILL_REG_3]]
-; GFX6: v_mov_b32_e32 [[OFFSET_REG3:v[0-9]+]], 0x[[OFFSET3:[0-9a-f]+]]
-; GFX6: buffer_store_dword [[SPILL_REG_3]], [[OFFSET_REG3]], s{{\[[0-9]+:[0-9]+\]}}, 0 offen
-; GFX6: buffer_load_dword [[SPILL_REG_3]], off, s{{\[[0-9]+:[0-9]+\]}}, 0{{$}}
-; GFX6: s_mov_b64 exec, s
-
-
-; GFX6: s_mov_b64 exec, 0xff
-; GFX6: buffer_store_dword [[SPILL_REG_4:v[0-9]+]]
-; GFX6-COUNT-4: v_writelane_b32 [[SPILL_REG_4]]
-; GFX6: v_mov_b32_e32 [[OFFSET_REG4:v[0-9]+]], 0x[[OFFSET4:[0-9a-f]+]]
-; GFX6: buffer_store_dword [[SPILL_REG_4]], [[OFFSET_REG4]], s{{\[[0-9]+:[0-9]+\]}}, 0 offen
-; GFX6: buffer_load_dword [[SPILL_REG_4]], off, s{{\[[0-9]+:[0-9]+\]}}, 0{{$}}
-; GFX6: s_mov_b64 exec, s
-
-; GFX6: NumSgprs: 48
-; GFX6: ScratchSize: 8608
-
-; FLATSCR: s_movk_i32 [[SOFF1:s[0-9]+]], 0x
-; GFX9-FLATSCR: s_waitcnt vmcnt(0)
-; FLATSCR: scratch_store_dwordx4 off, v[{{[0-9:]+}}], [[SOFF1]] ; 16-byte Folded Spill
-; FLATSCR: s_movk_i32 [[SOFF2:s[0-9]+]], 0x
-; FLATSCR: scratch_load_dwordx4 v[{{[0-9:]+}}], off, [[SOFF2]] ; 16-byte Folded Reload
define amdgpu_kernel void @test_limited_sgpr(<64 x i32> addrspace(1)* %out, <64 x i32> addrspace(1)* %in) #0 {
+; GFX6-LABEL: test_limited_sgpr:
+; GFX6: ; %bb.0: ; %entry
+; GFX6-NEXT: s_mov_b32 s40, SCRATCH_RSRC_DWORD0
+; GFX6-NEXT: s_mov_b32 s41, SCRATCH_RSRC_DWORD1
+; GFX6-NEXT: s_mov_b32 s42, -1
+; GFX6-NEXT: s_mov_b32 s43, 0xe8f000
+; GFX6-NEXT: s_add_u32 s40, s40, s3
+; GFX6-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GFX6-NEXT: v_mbcnt_lo_u32_b32_e64 v0, -1, 0
+; GFX6-NEXT: v_mbcnt_hi_u32_b32_e32 v5, -1, v0
+; GFX6-NEXT: v_mov_b32_e32 v6, 0
+; GFX6-NEXT: s_mov_b32 s38, 0
+; GFX6-NEXT: s_mov_b32 s39, 0xf000
+; GFX6-NEXT: s_waitcnt lgkmcnt(0)
+; GFX6-NEXT: s_mov_b64 s[36:37], s[2:3]
+; GFX6-NEXT: v_lshlrev_b32_e32 v7, 8, v5
+; GFX6-NEXT: v_mov_b32_e32 v8, v6
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[7:8], s[36:39], 0 addr64 offset:240
+; GFX6-NEXT: s_addc_u32 s41, s41, 0
+; GFX6-NEXT: s_mov_b32 s2, 0x83400
+; GFX6-NEXT: s_mov_b64 s[44:45], exec
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[40:43], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[40:43], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[40:43], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[40:43], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[7:8], s[36:39], 0 addr64 offset:224
+; GFX6-NEXT: s_mov_b32 s2, 0x83000
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[40:43], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[40:43], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[40:43], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[40:43], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[7:8], s[36:39], 0 addr64 offset:208
+; GFX6-NEXT: s_mov_b32 s2, 0x82c00
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[40:43], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[40:43], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[40:43], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[40:43], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[7:8], s[36:39], 0 addr64 offset:192
+; GFX6-NEXT: s_mov_b32 s2, 0x82800
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[40:43], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[40:43], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[40:43], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[40:43], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[7:8], s[36:39], 0 addr64 offset:176
+; GFX6-NEXT: s_mov_b32 s2, 0x82400
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[40:43], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[40:43], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[40:43], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[40:43], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[7:8], s[36:39], 0 addr64 offset:160
+; GFX6-NEXT: s_mov_b32 s2, 0x82000
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[40:43], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[40:43], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[40:43], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[40:43], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[7:8], s[36:39], 0 addr64 offset:144
+; GFX6-NEXT: s_mov_b32 s2, 0x81c00
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[40:43], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[40:43], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[40:43], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[40:43], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[7:8], s[36:39], 0 addr64 offset:128
+; GFX6-NEXT: s_mov_b32 s2, 0x81800
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[40:43], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[40:43], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[40:43], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[40:43], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[7:8], s[36:39], 0 addr64 offset:112
+; GFX6-NEXT: s_mov_b32 s2, 0x81400
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[40:43], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[40:43], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[40:43], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[40:43], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[7:8], s[36:39], 0 addr64 offset:96
+; GFX6-NEXT: s_mov_b32 s2, 0x81000
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[40:43], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[40:43], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[40:43], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[40:43], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[7:8], s[36:39], 0 addr64 offset:80
+; GFX6-NEXT: s_mov_b32 s2, 0x80c00
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[40:43], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[40:43], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[40:43], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[40:43], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[7:8], s[36:39], 0 addr64 offset:64
+; GFX6-NEXT: s_mov_b32 s2, 0x80400
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[40:43], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[40:43], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[40:43], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[40:43], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dwordx4 v[0:3], v[7:8], s[36:39], 0 addr64
+; GFX6-NEXT: buffer_load_dwordx4 v[9:12], v[7:8], s[36:39], 0 addr64 offset:16
+; GFX6-NEXT: s_mov_b32 s2, 0x80800
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v9, off, s[40:43], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v10, off, s[40:43], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v11, off, s[40:43], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v12, off, s[40:43], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_load_dwordx4 v[13:16], v[7:8], s[36:39], 0 addr64 offset:32
+; GFX6-NEXT: buffer_load_dwordx4 v[17:20], v[7:8], s[36:39], 0 addr64 offset:48
+; GFX6-NEXT: v_lshlrev_b32_e32 v4, 13, v0
+; GFX6-NEXT: v_add_i32_e32 v4, vcc, 16, v4
+; GFX6-NEXT: v_mov_b32_e32 v7, 1
+; GFX6-NEXT: buffer_store_dword v7, v4, s[40:43], 0 offen
+; GFX6-NEXT: ;;#ASMSTART
+; GFX6-NEXT: ; def s[4:11]
+; GFX6-NEXT: ;;#ASMEND
+; GFX6-NEXT: s_mov_b64 exec, 0xff
+; GFX6-NEXT: buffer_store_dword v4, off, s[40:43], 0
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: v_writelane_b32 v4, s4, 0
+; GFX6-NEXT: v_writelane_b32 v4, s5, 1
+; GFX6-NEXT: v_writelane_b32 v4, s6, 2
+; GFX6-NEXT: v_writelane_b32 v4, s7, 3
+; GFX6-NEXT: v_writelane_b32 v4, s8, 4
+; GFX6-NEXT: v_writelane_b32 v4, s9, 5
+; GFX6-NEXT: v_writelane_b32 v4, s10, 6
+; GFX6-NEXT: v_writelane_b32 v4, s11, 7
+; GFX6-NEXT: s_mov_b32 s2, 0x83800
+; GFX6-NEXT: buffer_store_dword v4, off, s[40:43], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v4, off, s[40:43], 0
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: s_mov_b64 exec, s[44:45]
+; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX6-NEXT: ;;#ASMSTART
+; GFX6-NEXT: ; def s[8:15]
+; GFX6-NEXT: ;;#ASMEND
+; GFX6-NEXT: ;;#ASMSTART
+; GFX6-NEXT: ; def s[16:23]
+; GFX6-NEXT: ;;#ASMEND
+; GFX6-NEXT: ;;#ASMSTART
+; GFX6-NEXT: ; def s[24:31]
+; GFX6-NEXT: ;;#ASMEND
+; GFX6-NEXT: ;;#ASMSTART
+; GFX6-NEXT: ; def s[4:7]
+; GFX6-NEXT: ;;#ASMEND
+; GFX6-NEXT: ;;#ASMSTART
+; GFX6-NEXT: ; def s[2:3]
+; GFX6-NEXT: ;;#ASMEND
+; GFX6-NEXT: ;;#ASMSTART
+; GFX6-NEXT: ; def s[36:37]
+; GFX6-NEXT: ;;#ASMEND
+; GFX6-NEXT: ;;#ASMSTART
+; GFX6-NEXT: ; def s33
+; GFX6-NEXT: ;;#ASMEND
+; GFX6-NEXT: s_and_saveexec_b64 s[34:35], vcc
+; GFX6-NEXT: s_cbranch_execz .LBB1_2
+; GFX6-NEXT: ; %bb.1: ; %bb0
+; GFX6-NEXT: s_mov_b64 s[44:45], exec
+; GFX6-NEXT: s_mov_b64 exec, 0xff
+; GFX6-NEXT: buffer_store_dword v9, off, s[40:43], 0
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: v_writelane_b32 v9, s8, 0
+; GFX6-NEXT: v_writelane_b32 v9, s9, 1
+; GFX6-NEXT: v_writelane_b32 v9, s10, 2
+; GFX6-NEXT: v_writelane_b32 v9, s11, 3
+; GFX6-NEXT: v_writelane_b32 v9, s12, 4
+; GFX6-NEXT: v_writelane_b32 v9, s13, 5
+; GFX6-NEXT: v_writelane_b32 v9, s14, 6
+; GFX6-NEXT: v_writelane_b32 v9, s15, 7
+; GFX6-NEXT: v_mov_b32_e32 v4, 0x2100
+; GFX6-NEXT: buffer_store_dword v9, v4, s[40:43], 0 offen ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[40:43], 0
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: s_mov_b64 exec, s[44:45]
+; GFX6-NEXT: s_mov_b64 s[44:45], exec
+; GFX6-NEXT: s_mov_b64 exec, 0xff
+; GFX6-NEXT: v_mov_b32_e32 v4, 0x20e0
+; GFX6-NEXT: buffer_store_dword v8, off, s[40:43], 0
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v8, v4, s[40:43], 0 offen ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: v_readlane_b32 s8, v8, 0
+; GFX6-NEXT: v_readlane_b32 s9, v8, 1
+; GFX6-NEXT: v_readlane_b32 s10, v8, 2
+; GFX6-NEXT: v_readlane_b32 s11, v8, 3
+; GFX6-NEXT: v_readlane_b32 s12, v8, 4
+; GFX6-NEXT: v_readlane_b32 s13, v8, 5
+; GFX6-NEXT: v_readlane_b32 s14, v8, 6
+; GFX6-NEXT: v_readlane_b32 s15, v8, 7
+; GFX6-NEXT: buffer_load_dword v8, off, s[40:43], 0
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: s_mov_b64 exec, s[44:45]
+; GFX6-NEXT: s_mov_b64 s[44:45], exec
+; GFX6-NEXT: s_mov_b64 exec, 0xff
+; GFX6-NEXT: buffer_store_dword v7, off, s[40:43], 0
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: v_writelane_b32 v7, s16, 0
+; GFX6-NEXT: v_writelane_b32 v7, s17, 1
+; GFX6-NEXT: v_writelane_b32 v7, s18, 2
+; GFX6-NEXT: v_writelane_b32 v7, s19, 3
+; GFX6-NEXT: v_writelane_b32 v7, s20, 4
+; GFX6-NEXT: v_writelane_b32 v7, s21, 5
+; GFX6-NEXT: v_writelane_b32 v7, s22, 6
+; GFX6-NEXT: v_writelane_b32 v7, s23, 7
+; GFX6-NEXT: v_mov_b32_e32 v4, 0x2120
+; GFX6-NEXT: buffer_store_dword v7, v4, s[40:43], 0 offen ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v7, off, s[40:43], 0
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: s_mov_b64 exec, s[44:45]
+; GFX6-NEXT: s_mov_b64 s[44:45], exec
+; GFX6-NEXT: s_mov_b64 exec, 0xff
+; GFX6-NEXT: v_mov_b32_e32 v4, 0x2100
+; GFX6-NEXT: buffer_store_dword v9, off, s[40:43], 0
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, v4, s[40:43], 0 offen ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: v_readlane_b32 s16, v9, 0
+; GFX6-NEXT: v_readlane_b32 s17, v9, 1
+; GFX6-NEXT: v_readlane_b32 s18, v9, 2
+; GFX6-NEXT: v_readlane_b32 s19, v9, 3
+; GFX6-NEXT: v_readlane_b32 s20, v9, 4
+; GFX6-NEXT: v_readlane_b32 s21, v9, 5
+; GFX6-NEXT: v_readlane_b32 s22, v9, 6
+; GFX6-NEXT: v_readlane_b32 s23, v9, 7
+; GFX6-NEXT: buffer_load_dword v9, off, s[40:43], 0
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: s_mov_b64 exec, s[44:45]
+; GFX6-NEXT: s_mov_b64 s[44:45], exec
+; GFX6-NEXT: s_mov_b64 exec, 0xff
+; GFX6-NEXT: buffer_store_dword v8, off, s[40:43], 0
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: v_writelane_b32 v8, s24, 0
+; GFX6-NEXT: v_writelane_b32 v8, s25, 1
+; GFX6-NEXT: v_writelane_b32 v8, s26, 2
+; GFX6-NEXT: v_writelane_b32 v8, s27, 3
+; GFX6-NEXT: v_writelane_b32 v8, s28, 4
+; GFX6-NEXT: v_writelane_b32 v8, s29, 5
+; GFX6-NEXT: v_writelane_b32 v8, s30, 6
+; GFX6-NEXT: v_writelane_b32 v8, s31, 7
+; GFX6-NEXT: v_mov_b32_e32 v4, 0x2140
+; GFX6-NEXT: buffer_store_dword v8, v4, s[40:43], 0 offen ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v8, off, s[40:43], 0
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: s_mov_b64 exec, s[44:45]
+; GFX6-NEXT: s_mov_b64 s[44:45], exec
+; GFX6-NEXT: s_mov_b64 exec, 0xff
+; GFX6-NEXT: v_mov_b32_e32 v4, 0x2120
+; GFX6-NEXT: buffer_store_dword v7, off, s[40:43], 0
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v7, v4, s[40:43], 0 offen ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: v_readlane_b32 s24, v7, 0
+; GFX6-NEXT: v_readlane_b32 s25, v7, 1
+; GFX6-NEXT: v_readlane_b32 s26, v7, 2
+; GFX6-NEXT: v_readlane_b32 s27, v7, 3
+; GFX6-NEXT: v_readlane_b32 s28, v7, 4
+; GFX6-NEXT: v_readlane_b32 s29, v7, 5
+; GFX6-NEXT: v_readlane_b32 s30, v7, 6
+; GFX6-NEXT: v_readlane_b32 s31, v7, 7
+; GFX6-NEXT: buffer_load_dword v7, off, s[40:43], 0
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: s_mov_b64 exec, s[44:45]
+; GFX6-NEXT: s_mov_b64 s[44:45], exec
+; GFX6-NEXT: s_mov_b64 exec, 15
+; GFX6-NEXT: buffer_store_dword v10, off, s[40:43], 0
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: v_writelane_b32 v10, s0, 0
+; GFX6-NEXT: v_writelane_b32 v10, s1, 1
+; GFX6-NEXT: v_writelane_b32 v10, s2, 2
+; GFX6-NEXT: v_writelane_b32 v10, s3, 3
+; GFX6-NEXT: v_mov_b32_e32 v4, 0x2160
+; GFX6-NEXT: buffer_store_dword v10, v4, s[40:43], 0 offen ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v10, off, s[40:43], 0
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: s_mov_b64 exec, s[44:45]
+; GFX6-NEXT: s_mov_b64 s[44:45], exec
+; GFX6-NEXT: s_mov_b64 exec, 15
+; GFX6-NEXT: buffer_store_dword v8, off, s[40:43], 0
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: v_writelane_b32 v8, s4, 0
+; GFX6-NEXT: v_writelane_b32 v8, s5, 1
+; GFX6-NEXT: v_writelane_b32 v8, s6, 2
+; GFX6-NEXT: v_writelane_b32 v8, s7, 3
+; GFX6-NEXT: s_mov_b32 s0, 0x85c00
+; GFX6-NEXT: buffer_store_dword v8, off, s[40:43], s0 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v8, off, s[40:43], 0
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: s_mov_b64 exec, s[44:45]
+; GFX6-NEXT: s_mov_b64 s[0:1], exec
+; GFX6-NEXT: s_mov_b64 exec, 3
+; GFX6-NEXT: buffer_store_dword v9, off, s[40:43], 0
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: v_writelane_b32 v9, s2, 0
+; GFX6-NEXT: v_writelane_b32 v9, s3, 1
+; GFX6-NEXT: s_mov_b32 s4, 0x86600
+; GFX6-NEXT: buffer_store_dword v9, off, s[40:43], s4 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, off, s[40:43], 0
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: s_mov_b64 exec, s[0:1]
+; GFX6-NEXT: s_mov_b64 s[44:45], exec
+; GFX6-NEXT: s_mov_b64 exec, 0xff
+; GFX6-NEXT: v_mov_b32_e32 v4, 0x2140
+; GFX6-NEXT: buffer_store_dword v7, off, s[40:43], 0
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v7, v4, s[40:43], 0 offen ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: v_readlane_b32 s0, v7, 0
+; GFX6-NEXT: v_readlane_b32 s1, v7, 1
+; GFX6-NEXT: v_readlane_b32 s2, v7, 2
+; GFX6-NEXT: v_readlane_b32 s3, v7, 3
+; GFX6-NEXT: v_readlane_b32 s4, v7, 4
+; GFX6-NEXT: v_readlane_b32 s5, v7, 5
+; GFX6-NEXT: v_readlane_b32 s6, v7, 6
+; GFX6-NEXT: v_readlane_b32 s7, v7, 7
+; GFX6-NEXT: buffer_load_dword v7, off, s[40:43], 0
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: s_mov_b64 exec, s[44:45]
+; GFX6-NEXT: s_mov_b64 s[44:45], exec
+; GFX6-NEXT: s_mov_b64 exec, 15
+; GFX6-NEXT: buffer_store_dword v8, off, s[40:43], 0
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: v_writelane_b32 v8, s36, 0
+; GFX6-NEXT: v_writelane_b32 v8, s37, 1
+; GFX6-NEXT: v_writelane_b32 v8, s38, 2
+; GFX6-NEXT: v_writelane_b32 v8, s39, 3
+; GFX6-NEXT: v_mov_b32_e32 v4, 0x2180
+; GFX6-NEXT: buffer_store_dword v8, v4, s[40:43], 0 offen ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v8, off, s[40:43], 0
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: s_mov_b64 exec, s[44:45]
+; GFX6-NEXT: s_mov_b64 s[38:39], exec
+; GFX6-NEXT: s_mov_b64 exec, 3
+; GFX6-NEXT: buffer_store_dword v10, off, s[40:43], 0
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: v_writelane_b32 v10, s36, 0
+; GFX6-NEXT: v_writelane_b32 v10, s37, 1
+; GFX6-NEXT: s_mov_b32 s44, 0x86400
+; GFX6-NEXT: buffer_store_dword v10, off, s[40:43], s44 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v10, off, s[40:43], 0
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: s_mov_b64 exec, s[38:39]
+; GFX6-NEXT: s_mov_b64 s[44:45], exec
+; GFX6-NEXT: s_mov_b64 exec, 15
+; GFX6-NEXT: v_mov_b32_e32 v4, 0x2170
+; GFX6-NEXT: buffer_store_dword v9, off, s[40:43], 0
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v9, v4, s[40:43], 0 offen ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: v_readlane_b32 s36, v9, 0
+; GFX6-NEXT: v_readlane_b32 s37, v9, 1
+; GFX6-NEXT: v_readlane_b32 s38, v9, 2
+; GFX6-NEXT: v_readlane_b32 s39, v9, 3
+; GFX6-NEXT: buffer_load_dword v9, off, s[40:43], 0
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: s_mov_b64 exec, s[44:45]
+; GFX6-NEXT: s_not_b64 exec, exec
+; GFX6-NEXT: v_mov_b32_e32 v4, 0x2190
+; GFX6-NEXT: buffer_store_dword v7, off, s[40:43], 0
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v7, v4, s[40:43], 0 offen ; 4-byte Folded Reload
+; GFX6-NEXT: s_not_b64 exec, exec
+; GFX6-NEXT: v_mov_b32_e32 v4, 0x2190
+; GFX6-NEXT: buffer_load_dword v7, v4, s[40:43], 0 offen ; 4-byte Folded Reload
+; GFX6-NEXT: s_not_b64 exec, exec
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: v_readlane_b32 s44, v7, 0
+; GFX6-NEXT: v_readlane_b32 s45, v7, 1
+; GFX6-NEXT: buffer_load_dword v7, off, s[40:43], 0
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: s_not_b64 exec, exec
+; GFX6-NEXT: s_mov_b64 vcc, s[34:35]
+; GFX6-NEXT: s_not_b64 exec, exec
+; GFX6-NEXT: v_mov_b32_e32 v4, 0x2198
+; GFX6-NEXT: buffer_store_dword v8, off, s[40:43], 0
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v8, v4, s[40:43], 0 offen ; 4-byte Folded Reload
+; GFX6-NEXT: s_not_b64 exec, exec
+; GFX6-NEXT: v_mov_b32_e32 v4, 0x2198
+; GFX6-NEXT: buffer_load_dword v8, v4, s[40:43], 0 offen ; 4-byte Folded Reload
+; GFX6-NEXT: s_not_b64 exec, exec
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: v_readlane_b32 s34, v8, 0
+; GFX6-NEXT: v_readlane_b32 s35, v8, 1
+; GFX6-NEXT: buffer_load_dword v8, off, s[40:43], 0
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: s_not_b64 exec, exec
+; GFX6-NEXT: ;;#ASMSTART
+; GFX6-NEXT: ; use s[8:15],s[16:23],s[24:31],s[0:7],s[36:39],s[34:35],s[44:45]
+; GFX6-NEXT: ;;#ASMEND
+; GFX6-NEXT: s_mov_b64 s[34:35], vcc
+; GFX6-NEXT: s_mov_b64 s[8:9], exec
+; GFX6-NEXT: s_mov_b64 exec, 15
+; GFX6-NEXT: s_mov_b32 s0, 0x86000
+; GFX6-NEXT: buffer_store_dword v4, off, s[40:43], 0
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v4, off, s[40:43], s0 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: v_readlane_b32 s36, v4, 0
+; GFX6-NEXT: v_readlane_b32 s37, v4, 1
+; GFX6-NEXT: v_readlane_b32 s38, v4, 2
+; GFX6-NEXT: v_readlane_b32 s39, v4, 3
+; GFX6-NEXT: buffer_load_dword v4, off, s[40:43], 0
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: s_mov_b64 exec, s[8:9]
+; GFX6-NEXT: s_mov_b64 s[4:5], exec
+; GFX6-NEXT: s_mov_b64 exec, 15
+; GFX6-NEXT: s_mov_b32 s6, 0x85800
+; GFX6-NEXT: buffer_store_dword v7, off, s[40:43], 0
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v7, off, s[40:43], s6 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: v_readlane_b32 s0, v7, 0
+; GFX6-NEXT: v_readlane_b32 s1, v7, 1
+; GFX6-NEXT: v_readlane_b32 s2, v7, 2
+; GFX6-NEXT: v_readlane_b32 s3, v7, 3
+; GFX6-NEXT: buffer_load_dword v7, off, s[40:43], 0
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: s_mov_b64 exec, s[4:5]
+; GFX6-NEXT: s_mov_b32 s2, 0x83800
+; GFX6-NEXT: buffer_store_dword v0, off, s[40:43], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v1, off, s[40:43], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v2, off, s[40:43], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v3, off, s[40:43], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_mov_b32 s2, 0x84000
+; GFX6-NEXT: buffer_store_dword v13, off, s[40:43], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v14, off, s[40:43], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v15, off, s[40:43], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v16, off, s[40:43], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_mov_b32 s2, 0x84800
+; GFX6-NEXT: buffer_store_dword v17, off, s[40:43], s2 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dword v18, off, s[40:43], s2 offset:4 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v19, off, s[40:43], s2 offset:8 ; 4-byte Folded Spill
+; GFX6-NEXT: buffer_store_dword v20, off, s[40:43], s2 offset:12 ; 4-byte Folded Spill
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: ;;#ASMSTART
+; GFX6-NEXT: ;;#ASMEND
+; GFX6-NEXT: s_mov_b32 s2, 0x84800
+; GFX6-NEXT: buffer_load_dword v17, off, s[40:43], s2 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v18, off, s[40:43], s2 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v19, off, s[40:43], s2 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v20, off, s[40:43], s2 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s2, 0x84000
+; GFX6-NEXT: buffer_load_dword v13, off, s[40:43], s2 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v14, off, s[40:43], s2 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v15, off, s[40:43], s2 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v16, off, s[40:43], s2 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s2, 0x83800
+; GFX6-NEXT: buffer_load_dword v0, off, s[40:43], s2 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v1, off, s[40:43], s2 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v2, off, s[40:43], s2 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v3, off, s[40:43], s2 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: ;;#ASMSTART
+; GFX6-NEXT: ;;#ASMEND
+; GFX6-NEXT: ;;#ASMSTART
+; GFX6-NEXT: ;;#ASMEND
+; GFX6-NEXT: ;;#ASMSTART
+; GFX6-NEXT: ;;#ASMEND
+; GFX6-NEXT: ;;#ASMSTART
+; GFX6-NEXT: ;;#ASMEND
+; GFX6-NEXT: ;;#ASMSTART
+; GFX6-NEXT: ;;#ASMEND
+; GFX6-NEXT: ;;#ASMSTART
+; GFX6-NEXT: ;;#ASMEND
+; GFX6-NEXT: .LBB1_2: ; %ret
+; GFX6-NEXT: s_or_b64 exec, exec, s[34:35]
+; GFX6-NEXT: s_mov_b32 s4, 0x83400
+; GFX6-NEXT: v_lshl_b64 v[4:5], v[5:6], 8
+; GFX6-NEXT: buffer_load_dword v6, off, s[40:43], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v7, off, s[40:43], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v8, off, s[40:43], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v9, off, s[40:43], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX6-NEXT: s_mov_b32 s4, 0x83000
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[6:9], v[4:5], s[0:3], 0 addr64 offset:240
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v6, off, s[40:43], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v7, off, s[40:43], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v8, off, s[40:43], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v9, off, s[40:43], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x82c00
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[6:9], v[4:5], s[0:3], 0 addr64 offset:224
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v6, off, s[40:43], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v7, off, s[40:43], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v8, off, s[40:43], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v9, off, s[40:43], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x82800
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[6:9], v[4:5], s[0:3], 0 addr64 offset:208
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v6, off, s[40:43], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v7, off, s[40:43], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v8, off, s[40:43], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v9, off, s[40:43], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x82400
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[6:9], v[4:5], s[0:3], 0 addr64 offset:192
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v6, off, s[40:43], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v7, off, s[40:43], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v8, off, s[40:43], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v9, off, s[40:43], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x82000
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[6:9], v[4:5], s[0:3], 0 addr64 offset:176
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v6, off, s[40:43], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v7, off, s[40:43], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v8, off, s[40:43], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v9, off, s[40:43], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x81c00
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[6:9], v[4:5], s[0:3], 0 addr64 offset:160
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v6, off, s[40:43], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v7, off, s[40:43], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v8, off, s[40:43], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v9, off, s[40:43], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x81800
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[6:9], v[4:5], s[0:3], 0 addr64 offset:144
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v6, off, s[40:43], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v7, off, s[40:43], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v8, off, s[40:43], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v9, off, s[40:43], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x81400
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[6:9], v[4:5], s[0:3], 0 addr64 offset:128
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v6, off, s[40:43], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v7, off, s[40:43], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v8, off, s[40:43], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v9, off, s[40:43], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x81000
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[6:9], v[4:5], s[0:3], 0 addr64 offset:112
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v6, off, s[40:43], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v7, off, s[40:43], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v8, off, s[40:43], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v9, off, s[40:43], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x80c00
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[6:9], v[4:5], s[0:3], 0 addr64 offset:96
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v6, off, s[40:43], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v7, off, s[40:43], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v8, off, s[40:43], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v9, off, s[40:43], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x80400
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[6:9], v[4:5], s[0:3], 0 addr64 offset:80
+; GFX6-NEXT: s_waitcnt expcnt(0)
+; GFX6-NEXT: buffer_load_dword v6, off, s[40:43], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v7, off, s[40:43], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v8, off, s[40:43], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v9, off, s[40:43], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_mov_b32 s4, 0x80800
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[6:9], v[4:5], s[0:3], 0 addr64 offset:64
+; GFX6-NEXT: buffer_store_dwordx4 v[17:20], v[4:5], s[0:3], 0 addr64 offset:48
+; GFX6-NEXT: buffer_store_dwordx4 v[13:16], v[4:5], s[0:3], 0 addr64 offset:32
+; GFX6-NEXT: s_waitcnt expcnt(2)
+; GFX6-NEXT: buffer_load_dword v6, off, s[40:43], s4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v7, off, s[40:43], s4 offset:4 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v8, off, s[40:43], s4 offset:8 ; 4-byte Folded Reload
+; GFX6-NEXT: buffer_load_dword v9, off, s[40:43], s4 offset:12 ; 4-byte Folded Reload
+; GFX6-NEXT: s_waitcnt vmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[6:9], v[4:5], s[0:3], 0 addr64 offset:16
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], v[4:5], s[0:3], 0 addr64
+; GFX6-NEXT: s_endpgm
+;
+; GFX9-FLATSCR-LABEL: test_limited_sgpr:
+; GFX9-FLATSCR: ; %bb.0: ; %entry
+; GFX9-FLATSCR-NEXT: s_load_dwordx4 s[36:39], s[0:1], 0x24
+; GFX9-FLATSCR-NEXT: v_mbcnt_lo_u32_b32 v0, -1, 0
+; GFX9-FLATSCR-NEXT: v_mbcnt_hi_u32_b32 v5, -1, v0
+; GFX9-FLATSCR-NEXT: v_lshlrev_b32_e32 v0, 8, v5
+; GFX9-FLATSCR-NEXT: s_add_u32 flat_scratch_lo, s2, s5
+; GFX9-FLATSCR-NEXT: s_waitcnt lgkmcnt(0)
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[1:4], v0, s[38:39] offset:240
+; GFX9-FLATSCR-NEXT: s_addc_u32 flat_scratch_hi, s3, 0
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x20b0
+; GFX9-FLATSCR-NEXT: v_mov_b32_e32 v6, 0
+; GFX9-FLATSCR-NEXT: v_mov_b32_e32 v7, 1
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[1:4], s0 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[8:11], v0, s[38:39] offset:224
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[1:4], v0, s[38:39] offset:208
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x20a0
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[1:4], s0 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[20:23], v0, s[38:39] offset:192
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[1:4], v0, s[38:39] offset:176
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x2090
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[1:4], s0 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[16:19], v0, s[38:39] offset:160
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[1:4], v0, s[38:39] offset:144
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x2080
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[1:4], s0 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[1:4], v0, s[38:39] offset:128
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x20c0
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[1:4], s0 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[1:4], v0, s[38:39] offset:112
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x2060
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[1:4], s0 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[1:4], v0, s[38:39] offset:96
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x2050
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[1:4], s0 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[1:4], v0, s[38:39] offset:80
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x2040
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[1:4], s0 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[1:4], v0, s[38:39] offset:64
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x2030
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[1:4], s0 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[1:4], v0, s[38:39] offset:48
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x2020
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[1:4], s0 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[1:4], v0, s[38:39] offset:32
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x2070
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[1:4], s0 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[1:4], v0, s[38:39] offset:16
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x2010
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[1:4], s0 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v0, s[38:39]
+; GFX9-FLATSCR-NEXT: v_mov_b32_e32 v4, 16
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: v_lshl_add_u32 v4, v0, 13, v4
+; GFX9-FLATSCR-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
+; GFX9-FLATSCR-NEXT: scratch_store_dword v4, v7, off
+; GFX9-FLATSCR-NEXT: ;;#ASMSTART
+; GFX9-FLATSCR-NEXT: ; def s[0:7]
+; GFX9-FLATSCR-NEXT: ;;#ASMEND
+; GFX9-FLATSCR-NEXT: ;;#ASMSTART
+; GFX9-FLATSCR-NEXT: ; def s[8:15]
+; GFX9-FLATSCR-NEXT: ;;#ASMEND
+; GFX9-FLATSCR-NEXT: ;;#ASMSTART
+; GFX9-FLATSCR-NEXT: ; def s[16:23]
+; GFX9-FLATSCR-NEXT: ;;#ASMEND
+; GFX9-FLATSCR-NEXT: ;;#ASMSTART
+; GFX9-FLATSCR-NEXT: ; def s[24:31]
+; GFX9-FLATSCR-NEXT: ;;#ASMEND
+; GFX9-FLATSCR-NEXT: ;;#ASMSTART
+; GFX9-FLATSCR-NEXT: ; def s[40:43]
+; GFX9-FLATSCR-NEXT: ;;#ASMEND
+; GFX9-FLATSCR-NEXT: ;;#ASMSTART
+; GFX9-FLATSCR-NEXT: ; def s[38:39]
+; GFX9-FLATSCR-NEXT: ;;#ASMEND
+; GFX9-FLATSCR-NEXT: ;;#ASMSTART
+; GFX9-FLATSCR-NEXT: ; def s[44:45]
+; GFX9-FLATSCR-NEXT: ;;#ASMEND
+; GFX9-FLATSCR-NEXT: ;;#ASMSTART
+; GFX9-FLATSCR-NEXT: ; def s33
+; GFX9-FLATSCR-NEXT: ;;#ASMEND
+; GFX9-FLATSCR-NEXT: s_and_saveexec_b64 s[34:35], vcc
+; GFX9-FLATSCR-NEXT: s_cbranch_execz .LBB1_2
+; GFX9-FLATSCR-NEXT: ; %bb.1: ; %bb0
+; GFX9-FLATSCR-NEXT: ;;#ASMSTART
+; GFX9-FLATSCR-NEXT: ; use s[0:7],s[8:15],s[16:23],s[24:31],s[40:43],s[38:39],s[44:45]
+; GFX9-FLATSCR-NEXT: ;;#ASMEND
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x20d0
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[0:3], s0 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x20e0
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[16:19], s0 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x20f0
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[20:23], s0 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x2100
+; GFX9-FLATSCR-NEXT: scratch_store_dwordx4 off, v[8:11], s0 ; 16-byte Folded Spill
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x2100
+; GFX9-FLATSCR-NEXT: ;;#ASMSTART
+; GFX9-FLATSCR-NEXT: ;;#ASMEND
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[8:11], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x20f0
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[20:23], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x20e0
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[16:19], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x20d0
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[0:3], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: ;;#ASMSTART
+; GFX9-FLATSCR-NEXT: ;;#ASMEND
+; GFX9-FLATSCR-NEXT: ;;#ASMSTART
+; GFX9-FLATSCR-NEXT: ;;#ASMEND
+; GFX9-FLATSCR-NEXT: ;;#ASMSTART
+; GFX9-FLATSCR-NEXT: ;;#ASMEND
+; GFX9-FLATSCR-NEXT: ;;#ASMSTART
+; GFX9-FLATSCR-NEXT: ;;#ASMEND
+; GFX9-FLATSCR-NEXT: ;;#ASMSTART
+; GFX9-FLATSCR-NEXT: ;;#ASMEND
+; GFX9-FLATSCR-NEXT: ;;#ASMSTART
+; GFX9-FLATSCR-NEXT: ;;#ASMEND
+; GFX9-FLATSCR-NEXT: .LBB1_2: ; %ret
+; GFX9-FLATSCR-NEXT: s_or_b64 exec, exec, s[34:35]
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x20b0
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[12:15], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: v_lshlrev_b64 v[4:5], 8, v[5:6]
+; GFX9-FLATSCR-NEXT: v_mov_b32_e32 v6, s37
+; GFX9-FLATSCR-NEXT: v_add_co_u32_e32 v4, vcc, s36, v4
+; GFX9-FLATSCR-NEXT: v_addc_co_u32_e32 v5, vcc, v6, v5, vcc
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x20a0
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[4:5], v[12:15], off offset:240
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[4:5], v[8:11], off offset:224
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[6:9], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x2090
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[4:5], v[6:9], off offset:208
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[4:5], v[20:23], off offset:192
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[20:23], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x2080
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[4:5], v[20:23], off offset:176
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[4:5], v[16:19], off offset:160
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[16:19], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x20c0
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[6:9], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x2060
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[12:15], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x2050
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(2)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[4:5], v[16:19], off offset:144
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(2)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[4:5], v[6:9], off offset:128
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[6:9], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x2040
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(3)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[4:5], v[12:15], off offset:112
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(1)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[4:5], v[6:9], off offset:96
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[6:9], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x2030
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[4:5], v[6:9], off offset:80
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[6:9], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x2020
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[4:5], v[6:9], off offset:64
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[6:9], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x2070
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[4:5], v[6:9], off offset:48
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[6:9], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_movk_i32 s0, 0x2010
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[4:5], v[6:9], off offset:32
+; GFX9-FLATSCR-NEXT: scratch_load_dwordx4 v[6:9], off, s0 ; 16-byte Folded Reload
+; GFX9-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[4:5], v[6:9], off offset:16
+; GFX9-FLATSCR-NEXT: global_store_dwordx4 v[4:5], v[0:3], off
+; GFX9-FLATSCR-NEXT: s_endpgm
+;
+; GFX10-FLATSCR-LABEL: test_limited_sgpr:
+; GFX10-FLATSCR: ; %bb.0: ; %entry
+; GFX10-FLATSCR-NEXT: s_add_u32 s2, s2, s5
+; GFX10-FLATSCR-NEXT: s_addc_u32 s3, s3, 0
+; GFX10-FLATSCR-NEXT: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s2
+; GFX10-FLATSCR-NEXT: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s3
+; GFX10-FLATSCR-NEXT: s_load_dwordx4 s[36:39], s[0:1], 0x24
+; GFX10-FLATSCR-NEXT: v_mbcnt_lo_u32_b32 v0, -1, 0
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v6, 0
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v7, 1
+; GFX10-FLATSCR-NEXT: s_mov_b32 s33, exec_lo
+; GFX10-FLATSCR-NEXT: v_mbcnt_hi_u32_b32 v5, -1, v0
+; GFX10-FLATSCR-NEXT: v_lshlrev_b32_e32 v0, 8, v5
+; GFX10-FLATSCR-NEXT: s_waitcnt lgkmcnt(0)
+; GFX10-FLATSCR-NEXT: s_clause 0xf
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[64:67], v0, s[38:39] offset:240
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[60:63], v0, s[38:39] offset:224
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[56:59], v0, s[38:39] offset:208
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[52:55], v0, s[38:39] offset:192
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[48:51], v0, s[38:39] offset:176
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[44:47], v0, s[38:39] offset:160
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[40:43], v0, s[38:39] offset:144
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[36:39], v0, s[38:39] offset:128
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[32:35], v0, s[38:39] offset:112
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[28:31], v0, s[38:39] offset:96
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[24:27], v0, s[38:39] offset:80
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[20:23], v0, s[38:39] offset:64
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[16:19], v0, s[38:39] offset:48
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[12:15], v0, s[38:39] offset:32
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[8:11], v0, s[38:39] offset:16
+; GFX10-FLATSCR-NEXT: global_load_dwordx4 v[0:3], v0, s[38:39]
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: v_lshl_add_u32 v4, v0, 13, 16
+; GFX10-FLATSCR-NEXT: scratch_store_dword v4, v7, off
+; GFX10-FLATSCR-NEXT: ;;#ASMSTART
+; GFX10-FLATSCR-NEXT: ; def s[0:7]
+; GFX10-FLATSCR-NEXT: ;;#ASMEND
+; GFX10-FLATSCR-NEXT: ;;#ASMSTART
+; GFX10-FLATSCR-NEXT: ; def s[8:15]
+; GFX10-FLATSCR-NEXT: ;;#ASMEND
+; GFX10-FLATSCR-NEXT: ;;#ASMSTART
+; GFX10-FLATSCR-NEXT: ; def s[16:23]
+; GFX10-FLATSCR-NEXT: ;;#ASMEND
+; GFX10-FLATSCR-NEXT: ;;#ASMSTART
+; GFX10-FLATSCR-NEXT: ; def s[24:31]
+; GFX10-FLATSCR-NEXT: ;;#ASMEND
+; GFX10-FLATSCR-NEXT: ;;#ASMSTART
+; GFX10-FLATSCR-NEXT: ; def s[40:43]
+; GFX10-FLATSCR-NEXT: ;;#ASMEND
+; GFX10-FLATSCR-NEXT: ;;#ASMSTART
+; GFX10-FLATSCR-NEXT: ; def s[34:35]
+; GFX10-FLATSCR-NEXT: ;;#ASMEND
+; GFX10-FLATSCR-NEXT: ;;#ASMSTART
+; GFX10-FLATSCR-NEXT: ; def s[38:39]
+; GFX10-FLATSCR-NEXT: ;;#ASMEND
+; GFX10-FLATSCR-NEXT: ;;#ASMSTART
+; GFX10-FLATSCR-NEXT: ; def s44
+; GFX10-FLATSCR-NEXT: ;;#ASMEND
+; GFX10-FLATSCR-NEXT: v_cmpx_eq_u32_e32 0, v0
+; GFX10-FLATSCR-NEXT: s_cbranch_execz .LBB1_2
+; GFX10-FLATSCR-NEXT: ; %bb.1: ; %bb0
+; GFX10-FLATSCR-NEXT: ;;#ASMSTART
+; GFX10-FLATSCR-NEXT: ; use s[0:7],s[8:15],s[16:23],s[24:31],s[40:43],s[34:35],s[38:39]
+; GFX10-FLATSCR-NEXT: ;;#ASMEND
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0x2010
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v88, v59
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v92, v63
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v87, v58
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v86, v57
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v85, v56
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v91, v62
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v90, v61
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v89, v60
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v60, v35
+; GFX10-FLATSCR-NEXT: scratch_store_dwordx4 off, v[64:67], s0 ; 16-byte Folded Spill
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v68, v39
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v59, v34
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v58, v33
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v57, v32
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v67, v38
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v66, v37
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v65, v36
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v36, v11
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v72, v43
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v76, v47
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v80, v51
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v84, v55
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v33, v8
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v71, v42
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v70, v41
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v69, v40
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v40, v15
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v75, v46
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v74, v45
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v73, v44
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v44, v19
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v79, v50
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v78, v49
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v77, v48
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v48, v23
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v83, v54
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v82, v53
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v81, v52
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v52, v27
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v56, v31
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v35, v10
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v34, v9
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v37, v12
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v41, v16
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v45, v20
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v49, v24
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v53, v28
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v39, v14
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v38, v13
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v43, v18
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v42, v17
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v47, v22
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v46, v21
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v51, v26
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v50, v25
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v55, v30
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v54, v29
+; GFX10-FLATSCR-NEXT: ;;#ASMSTART
+; GFX10-FLATSCR-NEXT: ;;#ASMEND
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v8, v33
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v28, v53
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v24, v49
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v20, v45
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v16, v41
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v12, v37
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v9, v34
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v10, v35
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v11, v36
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v32, v57
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v29, v54
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v30, v55
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v31, v56
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v25, v50
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v26, v51
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v27, v52
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v21, v46
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v22, v47
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v23, v48
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v17, v42
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v18, v43
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v19, v44
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v13, v38
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v14, v39
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v15, v40
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v33, v58
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v34, v59
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v35, v60
+; GFX10-FLATSCR-NEXT: ;;#ASMSTART
+; GFX10-FLATSCR-NEXT: ;;#ASMEND
+; GFX10-FLATSCR-NEXT: s_movk_i32 s0, 0x2010
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v36, v65
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v37, v66
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v38, v67
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v39, v68
+; GFX10-FLATSCR-NEXT: scratch_load_dwordx4 v[64:67], off, s0 ; 16-byte Folded Reload
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v60, v89
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v56, v85
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v52, v81
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v48, v77
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v44, v73
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v40, v69
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v61, v90
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v62, v91
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v63, v92
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v57, v86
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v58, v87
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v59, v88
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v53, v82
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v54, v83
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v55, v84
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v49, v78
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v50, v79
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v51, v80
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v45, v74
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v46, v75
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v47, v76
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v41, v70
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v42, v71
+; GFX10-FLATSCR-NEXT: v_mov_b32_e32 v43, v72
+; GFX10-FLATSCR-NEXT: ;;#ASMSTART
+; GFX10-FLATSCR-NEXT: ;;#ASMEND
+; GFX10-FLATSCR-NEXT: ;;#ASMSTART
+; GFX10-FLATSCR-NEXT: ;;#ASMEND
+; GFX10-FLATSCR-NEXT: ;;#ASMSTART
+; GFX10-FLATSCR-NEXT: ;;#ASMEND
+; GFX10-FLATSCR-NEXT: ;;#ASMSTART
+; GFX10-FLATSCR-NEXT: ;;#ASMEND
+; GFX10-FLATSCR-NEXT: ;;#ASMSTART
+; GFX10-FLATSCR-NEXT: ;;#ASMEND
+; GFX10-FLATSCR-NEXT: .LBB1_2: ; %ret
+; GFX10-FLATSCR-NEXT: s_or_b32 exec_lo, exec_lo, s33
+; GFX10-FLATSCR-NEXT: v_lshlrev_b64 v[4:5], 8, v[5:6]
+; GFX10-FLATSCR-NEXT: v_add_co_u32 v4, vcc_lo, s36, v4
+; GFX10-FLATSCR-NEXT: v_add_co_ci_u32_e32 v5, vcc_lo, s37, v5, vcc_lo
+; GFX10-FLATSCR-NEXT: s_waitcnt vmcnt(0)
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[4:5], v[64:67], off offset:240
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[4:5], v[60:63], off offset:224
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[4:5], v[56:59], off offset:208
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[4:5], v[52:55], off offset:192
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[4:5], v[48:51], off offset:176
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[4:5], v[44:47], off offset:160
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[4:5], v[40:43], off offset:144
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[4:5], v[36:39], off offset:128
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[4:5], v[32:35], off offset:112
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[4:5], v[28:31], off offset:96
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[4:5], v[24:27], off offset:80
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[4:5], v[20:23], off offset:64
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[4:5], v[16:19], off offset:48
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[4:5], v[12:15], off offset:32
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[4:5], v[8:11], off offset:16
+; GFX10-FLATSCR-NEXT: global_store_dwordx4 v[4:5], v[0:3], off
+; GFX10-FLATSCR-NEXT: s_endpgm
entry:
%lo = call i32 @llvm.amdgcn.mbcnt.lo(i32 -1, i32 0)
%tid = call i32 @llvm.amdgcn.mbcnt.hi(i32 -1, i32 %lo)
attributes #0 = { "amdgpu-waves-per-eu"="10,10" }
attributes #1 = { nounwind readnone }
+;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
+; CHECK: {{.*}}
+; FLATSCR: {{.*}}